Now Reading: Architecting Effective Decisions
-
01
Architecting Effective Decisions
00:00:00,000 –> 00:00:02,560
Every organization eventually gets the same request,
2
00:00:02,560 –> 00:00:04,360
put every KPI on one page.
3
00:00:04,360 –> 00:00:06,320
And everyone nods because it sounds reasonable.
4
00:00:06,320 –> 00:00:08,320
Executives want clarity, they want speed,
5
00:00:08,320 –> 00:00:10,280
they want to know what’s working and what’s failing
6
00:00:10,280 –> 00:00:11,960
without sitting through interpretive dance
7
00:00:11,960 –> 00:00:13,480
in a quarterly business review.
8
00:00:13,480 –> 00:00:15,360
But that request is a mistranslation.
9
00:00:15,360 –> 00:00:17,320
They aren’t asking for a prettier dashboard.
10
00:00:17,320 –> 00:00:20,480
They’re asking for a deterministic decision surface.
11
00:00:20,480 –> 00:00:22,680
One place where the organization can’t argue
12
00:00:22,680 –> 00:00:25,080
about definitions, can’t hide in nuance,
13
00:00:25,080 –> 00:00:29,480
and can’t delay action behind, we need to analyze it more.
14
00:00:29,480 –> 00:00:31,880
The reason the request keeps coming back is simple.
15
00:00:31,880 –> 00:00:34,000
The system leaks, meaning and trust.
16
00:00:34,000 –> 00:00:36,920
And when trust leaks, leaders ask for more visibility.
17
00:00:36,920 –> 00:00:38,200
Visibility won’t fix it.
18
00:00:38,200 –> 00:00:39,760
Decision architecture will.
19
00:00:39,760 –> 00:00:41,720
If you’re the person who has to defend decisions
20
00:00:41,720 –> 00:00:45,080
in audit meetings, post incident reviews, or board prep,
21
00:00:45,080 –> 00:00:47,200
you already know the dirty secret.
22
00:00:47,200 –> 00:00:49,280
We had a dashboard is not a control.
23
00:00:49,280 –> 00:00:50,160
It’s a screenshot.
24
00:00:50,160 –> 00:00:54,800
Subscribe if you want mental models and architectural patterns
25
00:00:54,800 –> 00:00:56,760
that survive contact with reality,
26
00:00:56,760 –> 00:00:59,400
governance, ownership, time constraints, and enforcement.
27
00:00:59,400 –> 00:01:01,960
Not feature tours, not click here.
28
00:01:01,960 –> 00:01:03,600
This show treats the Microsoft ecosystem
29
00:01:03,600 –> 00:01:04,560
like what it is.
30
00:01:04,560 –> 00:01:07,880
A distributed decision engine, you either design intentionally
31
00:01:07,880 –> 00:01:09,520
or you inherit accidentally.
32
00:01:09,520 –> 00:01:11,200
And inherited systems always drift.
33
00:01:11,200 –> 00:01:14,440
Now let’s name the real problem hiding behind the KPI request.
34
00:01:14,440 –> 00:01:18,960
The KPI request as a symptom, data entropy, in executive language.
35
00:01:18,960 –> 00:01:22,080
When an executive asks for all KPI’s on one page,
36
00:01:22,080 –> 00:01:24,400
the comfortable interpretation is that they’re impatient.
37
00:01:24,400 –> 00:01:26,560
The correct interpretation is that they’re responding
38
00:01:26,560 –> 00:01:28,720
to entropy, not thermodynamics.
39
00:01:28,720 –> 00:01:32,160
Enterprise entropy, drift, duplication, conflicting truths,
40
00:01:32,160 –> 00:01:35,080
and the slow breakdown of shared meaning across systems.
41
00:01:35,080 –> 00:01:38,280
KPI proliferation is the classic coping mechanism.
42
00:01:38,280 –> 00:01:41,120
When the organization can’t trust a small set of metrics,
43
00:01:41,120 –> 00:01:42,360
it starts collecting more.
44
00:01:42,360 –> 00:01:44,160
It adds a KPI for every argument.
45
00:01:44,160 –> 00:01:47,360
Every KPI becomes a political bandage over a system
46
00:01:47,360 –> 00:01:50,600
that can’t produce a single coherent answer without a meeting.
47
00:01:50,600 –> 00:01:53,400
And over time, more KPI’s turns into surveillance,
48
00:01:53,400 –> 00:01:55,200
not because leadership loves monitoring,
49
00:01:55,200 –> 00:01:56,480
because trust failed.
50
00:01:56,480 –> 00:01:59,800
And surveillance is what people buy when they can’t buy certainty.
51
00:01:59,800 –> 00:02:01,280
Here’s the thing most people miss.
52
00:02:01,280 –> 00:02:03,480
The KPI request isn’t about numbers.
53
00:02:03,480 –> 00:02:05,320
It’s about the cost of disagreement.
54
00:02:05,320 –> 00:02:07,480
A metric that requires interpretation isn’t a metric.
55
00:02:07,480 –> 00:02:09,960
It’s a conversation starter that sounds collaborative.
56
00:02:09,960 –> 00:02:11,720
It’s also why decisions take weeks.
57
00:02:11,720 –> 00:02:15,360
The organization has to negotiate reality every time it wants to act.
58
00:02:15,360 –> 00:02:18,520
This is why dashboards fail even when the data is accurate.
59
00:02:18,520 –> 00:02:20,320
Accuracy doesn’t create determinism.
60
00:02:20,320 –> 00:02:22,240
A perfectly accurate chart can still produce
61
00:02:22,240 –> 00:02:24,560
five different decisions depending on who’s looking at it,
62
00:02:24,560 –> 00:02:27,400
what incentives they have, and which definition of the metric
63
00:02:27,400 –> 00:02:29,000
they quietly prefer.
64
00:02:29,000 –> 00:02:32,800
That variance is entropy in its most expensive form, decision latency.
65
00:02:32,800 –> 00:02:35,360
Decision latency becomes the company’s real KPI,
66
00:02:35,360 –> 00:02:36,880
not because anyone tracks it,
67
00:02:36,880 –> 00:02:38,800
because it quietly dictates everything else.
68
00:02:38,800 –> 00:02:41,040
Mist targets, reactive cost-cutting,
69
00:02:41,040 –> 00:02:44,200
delayed escalations, incident breaches that were visible days earlier,
70
00:02:44,200 –> 00:02:45,440
but not acted on.
71
00:02:45,440 –> 00:02:46,960
You can’t out-report latency.
72
00:02:46,960 –> 00:02:48,120
You have to design it out.
73
00:02:48,120 –> 00:02:50,680
And this is where the one-page idea gets interesting.
74
00:02:50,680 –> 00:02:52,120
Executives don’t want a page.
75
00:02:52,120 –> 00:02:53,080
They want a control plane.
76
00:02:53,080 –> 00:02:57,160
They want one place where the organization’s operational story is consistent,
77
00:02:57,160 –> 00:02:59,520
and where the next action is not an emotional debate.
78
00:02:59,520 –> 00:03:02,920
So the request keeps returning because the system keeps leaking meaning.
79
00:03:02,920 –> 00:03:04,240
Let’s make the leak concrete.
80
00:03:04,240 –> 00:03:06,080
One team calculates revenue one way
81
00:03:06,080 –> 00:03:09,000
because their source system stores credits and returns differently.
82
00:03:09,000 –> 00:03:10,720
Another team calculates it another way
83
00:03:10,720 –> 00:03:12,480
because they normalize it downstream.
84
00:03:12,480 –> 00:03:14,320
Finance reconciles both in Excel
85
00:03:14,320 –> 00:03:17,200
because that’s where contradictions go to die quietly.
86
00:03:17,200 –> 00:03:19,240
Meanwhile, leadership is staring at a dashboard
87
00:03:19,240 –> 00:03:22,360
that is technically correct and strategically useless,
88
00:03:22,360 –> 00:03:26,040
because nobody can commit to action without first committing to a definition.
89
00:03:26,040 –> 00:03:28,840
That’s entropy, not missing data, competing truths.
90
00:03:28,840 –> 00:03:30,240
And it doesn’t stay in finance.
91
00:03:30,240 –> 00:03:32,600
I does the same thing with incidents in SLA’s.
92
00:03:32,600 –> 00:03:34,400
Severity gets negotiated after the fact.
93
00:03:34,400 –> 00:03:35,680
Tickets get reclassified.
94
00:03:35,680 –> 00:03:37,880
The clock gets paused by processed loopholes.
95
00:03:37,880 –> 00:03:40,640
The dashboard faithfully reports the outcome of the loopholes.
96
00:03:40,640 –> 00:03:42,480
Everybody claims the SLA is fine
97
00:03:42,480 –> 00:03:45,160
right up until a customer’s lawyer forces honesty.
98
00:03:45,160 –> 00:03:46,120
The system didn’t break.
99
00:03:46,120 –> 00:03:49,560
The system behaved like an ungoverned distributed environment.
100
00:03:49,560 –> 00:03:51,400
It optimized for local incentives,
101
00:03:51,400 –> 00:03:52,680
not global intent.
102
00:03:52,680 –> 00:03:54,280
This is the uncomfortable truth.
103
00:03:54,280 –> 00:03:56,040
Executives ask for more KPIs
104
00:03:56,040 –> 00:03:58,520
when they can’t trust the existing ones to trigger action
105
00:03:58,520 –> 00:04:01,240
because in a healthy system, a KPIs isn’t an observation
106
00:04:01,240 –> 00:04:04,320
but it’s an obligation it encodes when this happens we do that.
107
00:04:04,320 –> 00:04:07,640
If a KPIs doesn’t change behavior, it’s decoration with budget
108
00:04:07,640 –> 00:04:09,280
and decoration breeds more decoration.
109
00:04:09,280 –> 00:04:12,400
More tiles, more conditional formatting, more gradients,
110
00:04:12,400 –> 00:04:14,920
more can we add just one more metric.
111
00:04:14,920 –> 00:04:17,360
Until the page is full and nothing happens faster,
112
00:04:17,360 –> 00:04:19,600
you see it in Power BI projects all the time.
113
00:04:19,600 –> 00:04:22,480
Someone builds an overview page, cards, colors indicators
114
00:04:22,480 –> 00:04:24,160
because a video showed how to do it.
115
00:04:24,160 –> 00:04:26,880
The visual looks clean, everyone praises it.
116
00:04:26,880 –> 00:04:29,760
Then two weeks later, the same executive asks for the page again
117
00:04:29,760 –> 00:04:32,120
but simpler, clearer, more actionable.
118
00:04:32,120 –> 00:04:33,080
They’re not being difficult.
119
00:04:33,080 –> 00:04:35,280
They’re telling you the page didn’t reduce entropy.
120
00:04:35,280 –> 00:04:37,240
It just rendered entropy in high resolution.
121
00:04:37,240 –> 00:04:39,400
So treat the KPIs request as a diagnostic.
122
00:04:39,400 –> 00:04:41,840
It’s a symptom that the organization’s decision system
123
00:04:41,840 –> 00:04:45,000
is probabilistic, definitions drift, ownership is fuzzy,
124
00:04:45,000 –> 00:04:47,200
action pathways are negotiated in meetings
125
00:04:47,200 –> 00:04:49,160
and nothing is enforced by design.
126
00:04:49,160 –> 00:04:50,840
That’s what we’re fixing in this episode.
127
00:04:50,840 –> 00:04:52,720
But first you need a translation table
128
00:04:52,720 –> 00:04:54,720
because executives are actually pretty consistent.
129
00:04:54,720 –> 00:04:57,160
They just don’t speak architecture.
130
00:04:57,160 –> 00:04:59,400
What the boss says versus what the boss means,
131
00:04:59,400 –> 00:05:00,680
the translation table.
132
00:05:00,680 –> 00:05:03,680
Executives don’t ask for architecture because they don’t have to.
133
00:05:03,680 –> 00:05:05,840
They ask for outcomes, they ask for simplicity,
134
00:05:05,840 –> 00:05:07,120
they ask for a one-pager.
135
00:05:07,120 –> 00:05:08,840
Your job is to translate the request
136
00:05:08,840 –> 00:05:10,680
into the system requirement it implies,
137
00:05:10,680 –> 00:05:11,960
whether they know it or not.
138
00:05:11,960 –> 00:05:14,880
Because if you take the words literally, you’ll build a dashboard
139
00:05:14,880 –> 00:05:16,320
and you’ll get asked to rebuild it.
140
00:05:16,320 –> 00:05:19,480
So here’s the translation table, not as acute exercise
141
00:05:19,480 –> 00:05:20,920
as a survival mechanism.
142
00:05:20,920 –> 00:05:23,000
When the boss says I need everything on one page,
143
00:05:23,000 –> 00:05:25,200
they mean I need a single control plane.
144
00:05:25,200 –> 00:05:26,680
Not a layout.
145
00:05:26,680 –> 00:05:28,960
A control plane is where definitions are enforced,
146
00:05:28,960 –> 00:05:31,520
states are tracked, and escalation is predictable.
147
00:05:31,520 –> 00:05:32,880
A page is just an interface.
148
00:05:32,880 –> 00:05:35,040
An interface is lie with perfect confidence
149
00:05:35,040 –> 00:05:36,840
when the layers underneath disagree.
150
00:05:36,840 –> 00:05:39,080
If you give them nine tiles and a slicer,
151
00:05:39,080 –> 00:05:40,240
you gave them visibility.
152
00:05:40,240 –> 00:05:41,640
You did not give them control.
153
00:05:41,640 –> 00:05:44,480
They will still need a meeting to decide what the tiles mean,
154
00:05:44,480 –> 00:05:47,120
whether they can be trusted, and who is on the hook
155
00:05:47,120 –> 00:05:48,240
when one turns red.
156
00:05:48,240 –> 00:05:50,960
When the boss says make it simple, they mean
157
00:05:50,960 –> 00:05:53,760
remove interpretive freedom from critical metrics.
158
00:05:53,760 –> 00:05:55,360
They are not asking for fewer visuals,
159
00:05:55,360 –> 00:05:58,360
they’re asking to stop paying the organization to argue.
160
00:05:58,360 –> 00:06:01,160
Simplicity in executive language means the system
161
00:06:01,160 –> 00:06:03,400
should not allow 10 competing definitions
162
00:06:03,400 –> 00:06:05,240
of revenue, margin, or SLA.
163
00:06:05,240 –> 00:06:07,680
It means the number should arrive with a contract.
164
00:06:07,680 –> 00:06:10,000
How it’s calculated, what inputs it uses,
165
00:06:10,000 –> 00:06:12,200
what it excludes, and what action it triggers
166
00:06:12,200 –> 00:06:13,720
when it crosses a threshold.
167
00:06:13,720 –> 00:06:15,600
Most teams respond to make it simple
168
00:06:15,600 –> 00:06:17,560
by hiding complexity behind tool tips.
169
00:06:17,560 –> 00:06:19,040
That is not simplicity.
170
00:06:19,040 –> 00:06:20,680
That is camouflage.
171
00:06:20,680 –> 00:06:22,760
When the boss says just show red and green,
172
00:06:22,760 –> 00:06:25,960
they mean encode, risk tolerance, and trigger conditions.
173
00:06:25,960 –> 00:06:27,360
Red and green isn’t about color.
174
00:06:27,360 –> 00:06:29,320
It’s about thresholds that the organization
175
00:06:29,320 –> 00:06:30,680
commits to in advance.
176
00:06:30,680 –> 00:06:32,040
If the threshold isn’t explicit,
177
00:06:32,040 –> 00:06:34,000
the color becomes performance art.
178
00:06:34,000 –> 00:06:36,640
Someone will always argue that it’s only slightly red
179
00:06:36,640 –> 00:06:39,960
or red, but explainable, or red because the data’s wrong.
180
00:06:39,960 –> 00:06:40,760
They don’t want red.
181
00:06:40,760 –> 00:06:43,840
They want if it’s red, the next action is already decided.
182
00:06:43,840 –> 00:06:45,560
That’s the determinism they’re reaching for,
183
00:06:45,560 –> 00:06:46,680
even if they can’t name it.
184
00:06:46,680 –> 00:06:48,880
When the boss says we need this for the board,
185
00:06:48,880 –> 00:06:51,560
they mean we need audit-ready logic and lineage.
186
00:06:51,560 –> 00:06:53,120
Boards don’t care about your gradients.
187
00:06:53,120 –> 00:06:55,760
Boards care about whether the numbers are defensible.
188
00:06:55,760 –> 00:06:57,760
And defensible doesn’t mean you can talk fast
189
00:06:57,760 –> 00:06:58,600
in a meeting.
190
00:06:58,600 –> 00:07:00,080
It means you can show where the data came from,
191
00:07:00,080 –> 00:07:02,200
who approved the definitions, what changed,
192
00:07:02,200 –> 00:07:03,640
and why the decision that followed
193
00:07:03,640 –> 00:07:05,520
was consistent with policy.
194
00:07:05,520 –> 00:07:07,040
A board packet is not a report.
195
00:07:07,040 –> 00:07:08,600
It’s a liability document.
196
00:07:08,600 –> 00:07:11,320
If the organization can’t explain why a KPI moved
197
00:07:11,320 –> 00:07:13,760
and what it did about it, the board doesn’t see insight.
198
00:07:13,760 –> 00:07:15,160
It sees unmanaged risk.
199
00:07:15,160 –> 00:07:17,240
When the boss says, can we drill down?
200
00:07:17,240 –> 00:07:18,760
They mean can we resolve disputes
201
00:07:18,760 –> 00:07:21,000
without starting a new analytics project?
202
00:07:21,000 –> 00:07:23,640
Drill down is usually a euphemism for mistrust.
203
00:07:23,640 –> 00:07:25,920
It’s the executive version of prove it.
204
00:07:25,920 –> 00:07:27,920
And it’s reasonable, but if the only way to prove it
205
00:07:27,920 –> 00:07:30,080
is to spin up another analyst to build another page,
206
00:07:30,080 –> 00:07:32,560
you’ve built a dependency, not a decision system.
207
00:07:32,560 –> 00:07:35,120
In a decision engine, drill down is not exploration.
208
00:07:35,120 –> 00:07:36,280
It’s traceability.
209
00:07:36,280 –> 00:07:38,240
Show the inputs that produce the decision,
210
00:07:38,240 –> 00:07:40,920
the constraints applied, and the state of the triggered action.
211
00:07:40,920 –> 00:07:43,040
That’s how you end debates, not by adding more charts.
212
00:07:43,040 –> 00:07:45,840
When the boss says, just get it done by Friday, they mean,
213
00:07:45,840 –> 00:07:47,120
I will accept technical debt
214
00:07:47,120 –> 00:07:49,240
as long as it creates the illusion of control.
215
00:07:49,240 –> 00:07:50,720
This is the most dangerous one,
216
00:07:50,720 –> 00:07:53,360
because it’s where good architects become accomplices.
217
00:07:53,360 –> 00:07:55,520
Friday deadlines produce one-page dashboards
218
00:07:55,520 –> 00:07:58,400
that look authoritative but are structurally unverifiable.
219
00:07:58,400 –> 00:08:00,240
They create the appearance of a control plane
220
00:08:00,240 –> 00:08:02,720
while adding another entropy generator underneath.
221
00:08:02,720 –> 00:08:05,920
A brittle semantic model, a manual refresh work around,
222
00:08:05,920 –> 00:08:08,360
an Excel reconciliation step nobody documents,
223
00:08:08,360 –> 00:08:11,000
and then the deadline passes the dashboard exists,
224
00:08:11,000 –> 00:08:13,360
and the organization pretends the problem is solved
225
00:08:13,360 –> 00:08:14,360
until it isn’t.
226
00:08:14,360 –> 00:08:16,960
Here’s the checkpoint.
227
00:08:16,960 –> 00:08:20,720
If you remember nothing else, remember this distinction.
228
00:08:20,720 –> 00:08:22,720
Executives speak in interface requests,
229
00:08:22,720 –> 00:08:24,120
but they want decision guarantees.
230
00:08:24,120 –> 00:08:26,480
They want the organization to behave consistently
231
00:08:26,480 –> 00:08:27,600
when conditions change.
232
00:08:27,600 –> 00:08:29,840
They want fewer meetings, whose only output is,
233
00:08:29,840 –> 00:08:31,200
will monitor it.
234
00:08:31,200 –> 00:08:32,960
They want escalation without politics,
235
00:08:32,960 –> 00:08:35,520
so the KPI-1 page request is not a design briefing.
236
00:08:35,520 –> 00:08:37,440
It’s an architectural indictment.
237
00:08:37,440 –> 00:08:40,000
And the next trap is the one everyone falls into.
238
00:08:40,000 –> 00:08:42,360
Confusing visibility with control.
239
00:08:42,360 –> 00:08:43,920
Visibility is not control.
240
00:08:43,920 –> 00:08:46,000
Dashboards as well paper charts.
241
00:08:46,000 –> 00:08:48,200
Dashboards are good at one thing, showing state.
242
00:08:48,200 –> 00:08:50,800
They are terrible at the thing executives actually need,
243
00:08:50,800 –> 00:08:54,480
forcing the organization to respond to that state consistently.
244
00:08:54,480 –> 00:08:56,840
That distinction matters because most companies confuse
245
00:08:56,840 –> 00:08:58,840
I can see it with I can control it.
246
00:08:58,840 –> 00:09:00,840
And they treat visibility like a control system
247
00:09:00,840 –> 00:09:01,960
when it’s just telemetry.
248
00:09:01,960 –> 00:09:03,520
And telemetry without enforcement
249
00:09:03,520 –> 00:09:04,880
doesn’t create reliability.
250
00:09:04,880 –> 00:09:06,800
It creates meetings.
251
00:09:06,800 –> 00:09:08,760
A report can tell you revenue is down.
252
00:09:08,760 –> 00:09:11,680
Forecast variance is widening and churn is creeping upward.
253
00:09:11,680 –> 00:09:12,800
Great, now what?
254
00:09:12,800 –> 00:09:15,320
If the what is a debate loop, who owns it,
255
00:09:15,320 –> 00:09:17,400
whether the metric is real, whether it’s seasonal,
256
00:09:17,400 –> 00:09:19,320
whether the segment filter is fair,
257
00:09:19,320 –> 00:09:21,080
then you didn’t build a KPI system.
258
00:09:21,080 –> 00:09:24,160
You built a discussion forum with better typography.
259
00:09:24,160 –> 00:09:26,120
This is why so many executive dashboards
260
00:09:26,120 –> 00:09:27,640
become wallpaper charts.
261
00:09:27,640 –> 00:09:29,680
They look authoritative, they feel like control,
262
00:09:29,680 –> 00:09:31,760
they create comfort, and then they sit there
263
00:09:31,760 –> 00:09:33,840
while nothing deterministically happens.
264
00:09:33,840 –> 00:09:35,720
The thing most people miss is that executives
265
00:09:35,720 –> 00:09:38,600
don’t have time to become analysts and they shouldn’t have to.
266
00:09:38,600 –> 00:09:40,880
But dashboards quietly demand exactly that.
267
00:09:40,880 –> 00:09:44,440
They demand interpretation, they demand context reconstruction,
268
00:09:44,440 –> 00:09:46,160
they demand an understanding of the model,
269
00:09:46,160 –> 00:09:48,680
the refresh cadence, the exclusions, and the caveats,
270
00:09:48,680 –> 00:09:50,840
usually delivered verbally by the one person in the room
271
00:09:50,840 –> 00:09:52,160
who knows the data.
272
00:09:52,160 –> 00:09:54,040
That person becomes a single point of failure.
273
00:09:54,040 –> 00:09:56,480
And the organization calls that data driven.
274
00:09:56,480 –> 00:09:57,880
Now here’s where most people mess up.
275
00:09:57,880 –> 00:09:59,720
They try to fix this with interactivity,
276
00:09:59,720 –> 00:10:01,400
more slicers, more drill through,
277
00:10:01,400 –> 00:10:04,480
more bookmarks, more explore the data yourself.
278
00:10:04,480 –> 00:10:06,520
Interactivity doesn’t reduce ambiguity,
279
00:10:06,520 –> 00:10:08,920
it often multiplies it because now every executive
280
00:10:08,920 –> 00:10:11,040
can generate their own version of reality.
281
00:10:11,040 –> 00:10:14,040
One person filters by region, another filters by product.
282
00:10:14,040 –> 00:10:15,480
A third switch is the date range.
283
00:10:15,480 –> 00:10:17,320
You get three screenshots in a team’s chat
284
00:10:17,320 –> 00:10:19,800
and a new argument about which one is the truth.
285
00:10:19,800 –> 00:10:21,800
Congratulations, you turned a static disagreement
286
00:10:21,800 –> 00:10:23,080
into a distributed one.
287
00:10:23,080 –> 00:10:26,120
In architectural terms, you converted a deterministic model,
288
00:10:26,120 –> 00:10:28,640
one chart, one view into a probabilistic model.
289
00:10:28,640 –> 00:10:31,760
Many views, many interpretations, no enforced conclusion.
290
00:10:31,760 –> 00:10:33,240
And the cost isn’t theoretical.
291
00:10:33,240 –> 00:10:35,080
The hidden tax is decision latency.
292
00:10:35,080 –> 00:10:38,120
Decision latency is the time between we can see the problem
293
00:10:38,120 –> 00:10:40,600
and the organization actually changes behavior.
294
00:10:40,600 –> 00:10:42,640
Dashboards tend to optimize the first half
295
00:10:42,640 –> 00:10:43,880
and ignore the second.
296
00:10:43,880 –> 00:10:45,880
They make problems visible sooner,
297
00:10:45,880 –> 00:10:47,800
but they don’t compress the response cycle
298
00:10:47,800 –> 00:10:50,320
unless you deliberately wire response into the system.
299
00:10:50,320 –> 00:10:52,400
So you end up with a company that is incredibly
300
00:10:52,400 –> 00:10:54,720
aware of its problems and still slow.
301
00:10:54,720 –> 00:10:57,120
That’s why leadership keeps pushing for one page.
302
00:10:57,120 –> 00:10:59,280
They believe a tighter interface will compress time
303
00:10:59,280 –> 00:11:01,680
to decision, but time to decision doesn’t come from layout.
304
00:11:01,680 –> 00:11:02,920
It comes from pre-commitment.
305
00:11:02,920 –> 00:11:04,400
The system has to answer these questions
306
00:11:04,400 –> 00:11:06,840
before the meeting starts.
307
00:11:06,840 –> 00:11:10,320
What condition counts as bad, precisely, who is accountable
308
00:11:10,320 –> 00:11:11,280
specifically?
309
00:11:11,280 –> 00:11:14,280
What action happens automatically or by obligation?
310
00:11:14,280 –> 00:11:16,680
What time window exists before the risk compounds?
311
00:11:16,680 –> 00:11:18,400
Where is the state of that action tracked?
312
00:11:18,400 –> 00:11:20,600
So it can’t vanish into chat messages.
313
00:11:20,600 –> 00:11:22,040
A dashboard answers none of those.
314
00:11:22,040 –> 00:11:23,760
A dashboard is silent on ownership.
315
00:11:23,760 –> 00:11:24,880
Silent on enforcement.
316
00:11:24,880 –> 00:11:25,800
Silent on time.
317
00:11:25,800 –> 00:11:27,760
Silent on whether anyone did anything.
318
00:11:27,760 –> 00:11:30,840
That’s why executives get addicted to red, green indicators.
319
00:11:30,840 –> 00:11:32,160
They think they’re simplifying.
320
00:11:32,160 –> 00:11:35,040
What they’re really doing is begging you to encode a rule.
321
00:11:35,040 –> 00:11:37,080
They want the system to stop asking them to decide
322
00:11:37,080 –> 00:11:38,080
from scratch every week.
323
00:11:38,080 –> 00:11:40,480
Because if red only means we should discuss it,
324
00:11:40,480 –> 00:11:41,960
then red is not a signal.
325
00:11:41,960 –> 00:11:43,400
It’s an invitation to stall.
326
00:11:43,400 –> 00:11:44,680
This is the core misconception.
327
00:11:44,680 –> 00:11:45,960
Visibility is observation.
328
00:11:45,960 –> 00:11:47,480
Control is obligation.
329
00:11:47,480 –> 00:11:49,600
And Microsoft’s ecosystem for all its branding
330
00:11:49,600 –> 00:11:50,920
doesn’t magically fix that.
331
00:11:50,920 –> 00:11:52,520
Power BI can visualize.
332
00:11:52,520 –> 00:11:53,600
Fabric can store.
333
00:11:53,600 –> 00:11:54,800
Per view can catalog.
334
00:11:54,800 –> 00:11:55,920
Copilot can talk.
335
00:11:55,920 –> 00:11:59,080
None of that creates control unless you design a decision path
336
00:11:59,080 –> 00:12:01,320
that is deterministic and to end.
337
00:12:01,320 –> 00:12:04,000
This is why we’ll just build a scorecard often fails.
338
00:12:04,000 –> 00:12:04,840
It’s still a report.
339
00:12:04,840 –> 00:12:06,320
It still ends in interpretation.
340
00:12:06,320 –> 00:12:09,000
It still terminates in humans deciding whether to care.
341
00:12:09,000 –> 00:12:10,000
It’s governance theater.
342
00:12:10,000 –> 00:12:12,640
You can point at it in a deck, but it doesn’t enforce intent
343
00:12:12,640 –> 00:12:13,320
at scale.
344
00:12:13,320 –> 00:12:15,440
So if you’re the architect, stop treating the dashboard
345
00:12:15,440 –> 00:12:16,640
as the deliverable.
346
00:12:16,640 –> 00:12:18,800
The deliverable is the decision system behind it.
347
00:12:18,800 –> 00:12:21,040
The dashboard becomes a thin interface on top of rules,
348
00:12:21,040 –> 00:12:22,560
ownership, state, and execution.
349
00:12:22,560 –> 00:12:23,680
It’s not the control plane.
350
00:12:23,680 –> 00:12:25,400
It’s the window into the control plane.
351
00:12:25,400 –> 00:12:27,640
Once you accept that, everything else clicks.
352
00:12:27,640 –> 00:12:28,920
Because now the question stops being,
353
00:12:28,920 –> 00:12:31,320
how do we show all KPIs on one page?
354
00:12:31,320 –> 00:12:32,320
And becomes?
355
00:12:32,320 –> 00:12:34,120
Which decisions must the organization make
356
00:12:34,120 –> 00:12:34,960
deterministically?
357
00:12:34,960 –> 00:12:36,720
And what system will enforce them?
358
00:12:36,720 –> 00:12:37,360
Key.
359
00:12:37,360 –> 00:12:39,040
The foundational misunderstanding.
360
00:12:39,040 –> 00:12:40,400
KPIs are not metrics.
361
00:12:40,400 –> 00:12:42,040
KPIs are decision rules.
362
00:12:42,040 –> 00:12:43,400
The foundational misunderstanding is
363
00:12:43,400 –> 00:12:45,840
that organizations think a KPIs are fancy metric.
364
00:12:45,840 –> 00:12:46,600
It isn’t.
365
00:12:46,600 –> 00:12:47,840
A metric is observation.
366
00:12:47,840 –> 00:12:49,000
It describes what happened.
367
00:12:49,000 –> 00:12:50,240
A KPIs obligation.
368
00:12:50,240 –> 00:12:51,800
It describes what must happen next.
369
00:12:51,800 –> 00:12:54,360
That distinction matters because the moment you treat KPIs
370
00:12:54,360 –> 00:12:57,280
as important metrics, you accidentally permit interpretation.
371
00:12:57,280 –> 00:12:59,160
And interpretation is where entropy breeds.
372
00:12:59,160 –> 00:13:01,800
Interpretation is where every stakeholder can look
373
00:13:01,800 –> 00:13:04,400
at the same number and negotiate a different meaning,
374
00:13:04,400 –> 00:13:06,560
a different excuse, and a different delay.
375
00:13:06,560 –> 00:13:08,240
Most dashboards are built on metrics.
376
00:13:08,240 –> 00:13:12,160
They show revenue, margin, pipeline, SLA, customer satisfaction,
377
00:13:12,160 –> 00:13:12,920
whatever.
378
00:13:12,920 –> 00:13:15,760
And then the organization asks humans to do the hard part,
379
00:13:15,760 –> 00:13:18,160
decide whether it matters, decide who owns it,
380
00:13:18,160 –> 00:13:20,120
decide what to do, and decide how fast.
381
00:13:20,120 –> 00:13:21,400
That is not a control system.
382
00:13:21,400 –> 00:13:23,680
That is a weekly ritual where you pay senior people
383
00:13:23,680 –> 00:13:24,720
to improvise.
384
00:13:24,720 –> 00:13:27,080
A KPI in architectural terms is a rule
385
00:13:27,080 –> 00:13:29,760
that binds observation to a decision surface.
386
00:13:29,760 –> 00:13:31,520
It says, when this condition is true,
387
00:13:31,520 –> 00:13:34,120
the organization is no longer allowed to just notice.
388
00:13:34,120 –> 00:13:37,360
It is required to act or to formally declare an exception.
389
00:13:37,360 –> 00:13:38,960
This clicked for him after watching how
390
00:13:38,960 –> 00:13:42,400
KPI overview pages get praised and then quietly ignored.
391
00:13:42,400 –> 00:13:43,720
The visuals weren’t the problem.
392
00:13:43,720 –> 00:13:45,560
The absence of enforcement was.
393
00:13:45,560 –> 00:13:47,920
A KPI without an action is not inside.
394
00:13:47,920 –> 00:13:49,280
It’s decoration with budget.
395
00:13:49,280 –> 00:13:51,480
It’s a card visual that makes people feel governed
396
00:13:51,480 –> 00:13:54,400
while the organization continues to behave, however it wants.
397
00:13:54,400 –> 00:13:57,840
And once you accept that, you can start diagnosing KPI systems,
398
00:13:57,840 –> 00:14:00,040
the same way you diagnose security systems
399
00:14:00,040 –> 00:14:03,120
by checking whether they actually change behavior under pressure.
400
00:14:03,120 –> 00:14:04,680
Here’s the uncomfortable test.
401
00:14:04,680 –> 00:14:06,520
If a KPI turns red and nothing happens
402
00:14:06,520 –> 00:14:08,520
until the next meeting, it’s not a KPI.
403
00:14:08,520 –> 00:14:09,680
It’s a mood indicator.
404
00:14:09,680 –> 00:14:11,880
If a KPI turns red and the first response is,
405
00:14:11,880 –> 00:14:12,840
is the data right?
406
00:14:12,840 –> 00:14:13,880
It’s not a KPI.
407
00:14:13,880 –> 00:14:16,240
It’s a trust problem, wearing a number as a costume.
408
00:14:16,240 –> 00:14:18,080
If a KPI turns red and five people
409
00:14:18,080 –> 00:14:20,080
can claim partial ownership, it’s not a KPI.
410
00:14:20,080 –> 00:14:21,200
It’s a political object.
411
00:14:21,200 –> 00:14:23,560
A real KPI is what executives think they’re
412
00:14:23,560 –> 00:14:25,560
buying when they ask for red and green.
413
00:14:25,560 –> 00:14:28,760
They think red means something is now in motion,
414
00:14:28,760 –> 00:14:30,520
not we will add it to the agenda.
415
00:14:30,520 –> 00:14:32,200
So redefine it properly.
416
00:14:32,200 –> 00:14:34,200
A metric answers, what is the value?
417
00:14:34,200 –> 00:14:37,240
A KPI answers, what are we going to do about that value?
418
00:14:37,240 –> 00:14:39,480
And who is obligated to do it and buy when?
419
00:14:39,480 –> 00:14:41,600
That’s why the same metric can exist for years
420
00:14:41,600 –> 00:14:44,240
without changing outcomes, then suddenly become transformative
421
00:14:44,240 –> 00:14:46,000
when you bind it to a decision rule.
422
00:14:46,000 –> 00:14:49,800
Nothing about the number changed, the obligation did.
423
00:14:49,800 –> 00:14:51,320
Now there’s a second trap, and it’s
424
00:14:51,320 –> 00:14:54,680
where naive KPI enforcement collapses into dysfunction, gaming.
425
00:14:54,680 –> 00:14:56,600
The moment you attach consequences to a number,
426
00:14:56,600 –> 00:14:58,320
people will optimize the number.
427
00:14:58,320 –> 00:15:00,520
Not the outcome, the system didn’t become broken.
428
00:15:00,520 –> 00:15:03,880
It behaved like an incentive-driven distributed environment.
429
00:15:03,880 –> 00:15:06,280
It took the path of least resistance.
430
00:15:06,280 –> 00:15:08,440
That’s why KPI design without counterbalances
431
00:15:08,440 –> 00:15:10,880
becomes a factory for perverse incentives.
432
00:15:10,880 –> 00:15:13,080
If you obsess over SLA compliance,
433
00:15:13,080 –> 00:15:15,960
teams will reclassify incidents to protect the clock.
434
00:15:15,960 –> 00:15:17,720
If you obsess over forecast accuracy,
435
00:15:17,720 –> 00:15:20,240
teams will sandbag the forecast so they can beat it.
436
00:15:20,240 –> 00:15:22,920
If you obsess over pipeline volume, you will get pipeline.
437
00:15:22,920 –> 00:15:24,040
It just won’t convert.
438
00:15:24,040 –> 00:15:25,840
So KPI’s decision rule doesn’t mean
439
00:15:25,840 –> 00:15:27,600
punish people with numbers.
440
00:15:27,600 –> 00:15:30,280
It means enforce intent with paired rules.
441
00:15:30,280 –> 00:15:32,440
One KPI for the target behavior,
442
00:15:32,440 –> 00:15:35,160
and one for the side effects you refuse to tolerate.
443
00:15:35,160 –> 00:15:37,720
Throughput paired with quality, speed paired with risk,
444
00:15:37,720 –> 00:15:39,800
growth paired with margin, automation paired
445
00:15:39,800 –> 00:15:42,600
with false positive rate, because you are not building a report,
446
00:15:42,600 –> 00:15:44,360
you are building a behavioral system.
447
00:15:44,360 –> 00:15:47,080
And systems drift toward whatever is easiest to satisfy
448
00:15:47,080 –> 00:15:49,520
unless you encode constraints that force the system
449
00:15:49,520 –> 00:15:50,240
to stay honest.
450
00:15:50,240 –> 00:15:53,120
That’s the entire point of deterministic decision making
451
00:15:53,120 –> 00:15:55,560
to stop renegotiating intent every week,
452
00:15:55,560 –> 00:15:57,680
to stop letting convenience override policy,
453
00:15:57,680 –> 00:16:00,120
to stop pretending that visibility is governance.
454
00:16:00,120 –> 00:16:01,960
So when someone says we need KPIs,
455
00:16:01,960 –> 00:16:04,120
the correct follow-up isn’t which metrics.
456
00:16:04,120 –> 00:16:07,400
It’s which decisions must become non-negotiable.
457
00:16:07,400 –> 00:16:09,400
Because that is what a KPI really is.
458
00:16:09,400 –> 00:16:12,320
A decision rule the organization commits to in advance,
459
00:16:12,320 –> 00:16:14,400
backed by ownership, time constraints,
460
00:16:14,400 –> 00:16:17,440
and an enforcement path that doesn’t require a meeting to exist.
461
00:16:17,440 –> 00:16:19,360
And once you accept that definition,
462
00:16:19,360 –> 00:16:21,680
the next question becomes inevitable.
463
00:16:21,680 –> 00:16:23,440
What are the minimum non-negotiables
464
00:16:23,440 –> 00:16:27,120
that KPI needs before you’re allowed to call it a KPI at all?
465
00:16:27,120 –> 00:16:30,120
The five non-negotiables of a deterministic decision engine.
466
00:16:30,120 –> 00:16:32,160
So here they are, the five non-negotiables,
467
00:16:32,160 –> 00:16:34,520
not best practices, not it depends.
468
00:16:34,520 –> 00:16:37,040
The minimum structure required before a KPI
469
00:16:37,040 –> 00:16:39,280
stops being a decorative metric and becomes
470
00:16:39,280 –> 00:16:40,800
a deterministic decision rule.
471
00:16:40,800 –> 00:16:43,520
And yes, this is the part where most organizations get uncomfortable
472
00:16:43,520 –> 00:16:46,080
because each non-negotiable removes a kind of freedom.
473
00:16:46,080 –> 00:16:47,520
And organizations love freedom.
474
00:16:47,520 –> 00:16:49,840
Right up until that freedom turns into drift, delay,
475
00:16:49,840 –> 00:16:51,560
and plausible deniability.
476
00:16:51,560 –> 00:16:53,680
Non-negotiable one is trigger definition.
477
00:16:53,680 –> 00:16:55,800
A trigger is not if it gets worse,
478
00:16:55,800 –> 00:16:58,600
or if it trends down, or my personal favorite,
479
00:16:58,600 –> 00:17:00,040
if we start to see a decline.
480
00:17:00,040 –> 00:17:00,880
That is not a trigger.
481
00:17:00,880 –> 00:17:01,960
That’s a future argument.
482
00:17:01,960 –> 00:17:03,520
A trigger is a precise condition.
483
00:17:03,520 –> 00:17:05,680
Threshold, duration, and context constraints.
484
00:17:05,680 –> 00:17:07,840
Threshold means you commit to a number.
485
00:17:07,840 –> 00:17:11,560
Duration means you commit to how long the number must persist before action.
486
00:17:11,560 –> 00:17:13,680
Context means you commit to scope,
487
00:17:13,680 –> 00:17:15,600
which segment, region, product line,
488
00:17:15,600 –> 00:17:17,640
severity, or customer tier qualifies.
489
00:17:17,640 –> 00:17:21,240
If forecast variance exceeds negative 7% for 10 business days
490
00:17:21,240 –> 00:17:23,040
in e-mail enterprise accounts.
491
00:17:23,040 –> 00:17:24,840
Now the system knows what bad means.
492
00:17:24,840 –> 00:17:27,640
More importantly, humans don’t get to redefine bad
493
00:17:27,640 –> 00:17:29,880
in the meeting to match the story they want to tell.
494
00:17:29,880 –> 00:17:31,920
Non-negotiable too is ownership lock,
495
00:17:31,920 –> 00:17:35,880
not finance owns revenue, not IT owns incidents.
496
00:17:35,880 –> 00:17:37,000
Those are departments.
497
00:17:37,000 –> 00:17:39,560
Departments are how accountability goes to die.
498
00:17:39,560 –> 00:17:42,840
A deterministic decision engine has one accountable role per trigger.
499
00:17:42,840 –> 00:17:45,000
One, a named function, not a distribution list.
500
00:17:45,000 –> 00:17:48,040
The VP of revenue operations, the incident commander role,
501
00:17:48,040 –> 00:17:50,240
the service owner for a specific product.
502
00:17:50,240 –> 00:17:52,720
An ownership is not responsible for reporting.
503
00:17:52,720 –> 00:17:55,800
Ownership means the trigger creates an obligation on that role
504
00:17:55,800 –> 00:17:57,800
to execute the pre-committed action,
505
00:17:57,800 –> 00:18:01,280
or to file an exception that is visible, time stamped, and explainable.
506
00:18:01,280 –> 00:18:03,480
If you can’t point to one accountable role,
507
00:18:03,480 –> 00:18:06,120
the system has already chosen politics over determinism.
508
00:18:06,120 –> 00:18:08,600
Non-negotiable three is pre-committed action.
509
00:18:08,600 –> 00:18:10,720
This is the one that exposes whether you’re serious.
510
00:18:10,720 –> 00:18:14,640
If a KPI turning red results in let schedule time to discuss,
511
00:18:14,640 –> 00:18:17,760
you have built a notification system, not a decision system.
512
00:18:17,760 –> 00:18:20,760
A decision engine has actions attached in advance.
513
00:18:20,760 –> 00:18:23,600
When the trigger fires, the next move is already defined.
514
00:18:23,600 –> 00:18:26,680
The meeting, if it happens at all, is for exception handling,
515
00:18:26,680 –> 00:18:28,600
not for inventing the response from scratch.
516
00:18:28,600 –> 00:18:32,240
For forecast variance, pre-committed actions might include
517
00:18:32,240 –> 00:18:35,360
freeze discretionary spend above a defined threshold,
518
00:18:35,360 –> 00:18:37,760
reallocate a defined portion of paid budget
519
00:18:37,760 –> 00:18:41,280
to the top performing channel, trigger a deal acceleration sprint
520
00:18:41,280 –> 00:18:44,080
for late stage pipeline, or force a forecast reset
521
00:18:44,080 –> 00:18:45,520
using a greed logic.
522
00:18:45,520 –> 00:18:49,080
For incident SLA risk, pre-committed actions might include
523
00:18:49,080 –> 00:18:51,720
escalate to on-call tier 2 after a fixed time,
524
00:18:51,720 –> 00:18:54,800
open a major incident bridge, automatically pre-stage customer
525
00:18:54,800 –> 00:18:57,920
communications, or trigger a known remediation playbook.
526
00:18:57,920 –> 00:18:59,400
And the action must be explicit enough
527
00:18:59,400 –> 00:19:01,320
that it can be executed consistently.
528
00:19:01,320 –> 00:19:02,920
If the action requires interpretation,
529
00:19:02,920 –> 00:19:04,440
you move the entropy downstream.
530
00:19:04,440 –> 00:19:05,720
You did not remove it.
531
00:19:05,720 –> 00:19:07,680
Non-negotiable for is time constraint.
532
00:19:07,680 –> 00:19:09,920
This is where dashboards quietly sabotage you,
533
00:19:09,920 –> 00:19:12,040
because dashboards align to meeting schedules
534
00:19:12,040 –> 00:19:14,040
and meeting schedules align to calendars
535
00:19:14,040 –> 00:19:15,720
and calendars don’t care about risk.
536
00:19:15,720 –> 00:19:19,000
A decision engine ties execution windows to the risk curve.
537
00:19:19,000 –> 00:19:21,560
If the system detects forecast variance compounding,
538
00:19:21,560 –> 00:19:23,880
the response window might be 48 hours,
539
00:19:23,880 –> 00:19:25,840
not by next month’s review.
540
00:19:25,840 –> 00:19:28,080
If the system detects SLA breach risk,
541
00:19:28,080 –> 00:19:30,000
the response window might be 15 minutes
542
00:19:30,000 –> 00:19:32,600
for severity 1, 1 hour for severity 2, and so on.
543
00:19:32,600 –> 00:19:35,200
Time constraints convert awareness into urgency
544
00:19:35,200 –> 00:19:38,000
without needing someone to manufacture urgency in a meeting.
545
00:19:38,000 –> 00:19:40,280
They also create a measurable property you can manage,
546
00:19:40,280 –> 00:19:42,040
compliance to the response window,
547
00:19:42,040 –> 00:19:44,520
because if you can’t act within the required time window,
548
00:19:44,520 –> 00:19:46,320
the trigger is meaningless theater.
549
00:19:46,320 –> 00:19:48,080
It’s telling you the building is on fire
550
00:19:48,080 –> 00:19:50,160
and scheduling a discussion for Tuesday,
551
00:19:50,160 –> 00:19:52,000
non-negotiable fire is feedback loop.
552
00:19:52,000 –> 00:19:54,680
Most KPI systems stop at triggering an escalation.
553
00:19:54,680 –> 00:19:55,840
They fire alarms.
554
00:19:55,840 –> 00:19:57,200
They root messages.
555
00:19:57,200 –> 00:19:58,760
They generate activity.
556
00:19:58,760 –> 00:20:00,280
Then they forget.
557
00:20:01,280 –> 00:20:03,640
A deterministic engine retains memory,
558
00:20:03,640 –> 00:20:05,680
what fired, what action executed,
559
00:20:05,680 –> 00:20:08,040
who approved or overroaded, how long it took,
560
00:20:08,040 –> 00:20:09,440
and what outcome followed.
561
00:20:09,440 –> 00:20:11,120
And then it measures intervention efficacy.
562
00:20:11,120 –> 00:20:12,600
Did the variance stabilize?
563
00:20:12,600 –> 00:20:14,160
Did forecast error reduce?
564
00:20:14,160 –> 00:20:17,080
Did SLA compliance return above the committed threshold?
565
00:20:17,080 –> 00:20:18,560
Did MTT are improved?
566
00:20:18,560 –> 00:20:20,120
Did false escalations increase?
567
00:20:20,120 –> 00:20:23,520
This is where you move from automation to control.
568
00:20:23,520 –> 00:20:25,720
Because now the organization can tune decisions
569
00:20:25,720 –> 00:20:27,440
the same way it tunes systems.
570
00:20:27,440 –> 00:20:30,000
By observing outcomes and adjusting rules deliberately,
571
00:20:30,000 –> 00:20:31,320
not by changing opinions.
572
00:20:31,320 –> 00:20:32,720
If you skip the feedback loop,
573
00:20:32,720 –> 00:20:34,240
you create brittle automation
574
00:20:34,240 –> 00:20:36,880
that slowly becomes wrong as the environment changes.
575
00:20:36,880 –> 00:20:38,000
The system doesn’t adapt.
576
00:20:38,000 –> 00:20:39,200
It accumulates exceptions.
577
00:20:39,200 –> 00:20:41,040
Exceptions become the new normal.
578
00:20:41,040 –> 00:20:42,200
An entropy wins again.
579
00:20:42,200 –> 00:20:43,520
So those are the five trigger,
580
00:20:43,520 –> 00:20:45,040
owner action time feedback.
581
00:20:45,040 –> 00:20:46,360
They are not optional knobs.
582
00:20:46,360 –> 00:20:48,800
They are the architecture that makes a KPI real.
583
00:20:48,800 –> 00:20:50,320
And once you enforce all five,
584
00:20:50,320 –> 00:20:52,920
you’ll notice something that executives can’t articulate
585
00:20:52,920 –> 00:20:54,000
but immediately feel.
586
00:20:54,000 –> 00:20:56,840
Meetings get shorter, debates get rarer,
587
00:20:56,840 –> 00:20:59,880
and the organization starts behaving like it has intent.
588
00:20:59,880 –> 00:21:00,960
Not just information.
589
00:21:00,960 –> 00:21:02,240
Now the next question is obvious.
590
00:21:02,240 –> 00:21:04,840
If this is the structure, what does it look like as a stack?
591
00:21:04,840 –> 00:21:06,280
Not as products, as layers.
592
00:21:06,280 –> 00:21:07,400
The decision stack.
593
00:21:07,400 –> 00:21:10,280
Data, logic, state, action, interface.
594
00:21:10,280 –> 00:21:12,560
Once you accept the five non-negotiables,
595
00:21:12,560 –> 00:21:15,440
you stop thinking in pages and start thinking in layers.
596
00:21:15,440 –> 00:21:17,600
Because determinism is not a visual property,
597
00:21:17,600 –> 00:21:19,280
it’s an architectural property.
598
00:21:19,280 –> 00:21:20,920
And the fastest way to explain it,
599
00:21:20,920 –> 00:21:23,280
especially in a Microsoft heavy organization,
600
00:21:23,280 –> 00:21:24,560
is as a decision stack.
601
00:21:24,560 –> 00:21:26,120
Not a tool chain, a stack,
602
00:21:26,120 –> 00:21:28,640
a sequence of layers where each layer has a job
603
00:21:28,640 –> 00:21:30,600
and each job removes a class of ambiguity
604
00:21:30,600 –> 00:21:32,400
the organization normally tolerates.
605
00:21:32,400 –> 00:21:36,840
Data, logic, dot state, dot action, interface.
606
00:21:36,840 –> 00:21:38,880
If any one of these layers is missing,
607
00:21:38,880 –> 00:21:42,120
the KPI system collapses back into reporting theater.
608
00:21:42,120 –> 00:21:43,880
Layer one is data.
609
00:21:43,880 –> 00:21:46,280
This is where most KPI conversations start,
610
00:21:46,280 –> 00:21:47,720
and it’s usually where they end.
611
00:21:47,720 –> 00:21:49,640
People fight about source systems,
612
00:21:49,640 –> 00:21:51,920
refresh cadence, data quality, and lineage.
613
00:21:51,920 –> 00:21:52,960
Important, yes.
614
00:21:52,960 –> 00:21:54,480
But here’s the uncomfortable truth.
615
00:21:54,480 –> 00:21:56,800
Even perfect data doesn’t produce a decision.
616
00:21:56,800 –> 00:21:58,520
It produces arguments faster.
617
00:21:58,520 –> 00:22:00,960
Data’s job in the stack is not to be available.
618
00:22:00,960 –> 00:22:03,160
Data’s job is to converge into something
619
00:22:03,160 –> 00:22:06,440
that the organization agrees is eligible to drive automation,
620
00:22:06,440 –> 00:22:08,120
eligible to trigger obligation,
621
00:22:08,120 –> 00:22:10,280
eligible to get someone woken up at 2am.
622
00:22:10,280 –> 00:22:12,240
If you can’t say that with a straight face,
623
00:22:12,240 –> 00:22:14,120
you don’t have decision-grade data.
624
00:22:14,120 –> 00:22:16,680
You have analytics-grade data, different category,
625
00:22:16,680 –> 00:22:19,240
different consequences, layer two is logic.
626
00:22:19,240 –> 00:22:20,880
This is where meaning gets compiled,
627
00:22:20,880 –> 00:22:25,360
not inferred, not interpreted by the business, compiled.
628
00:22:25,360 –> 00:22:28,120
A semantic model is basically an authorization compiler
629
00:22:28,120 –> 00:22:28,960
for meaning.
630
00:22:28,960 –> 00:22:31,960
It takes raw facts and turns them into governed definitions.
631
00:22:31,960 –> 00:22:34,840
The organization can reuse without relitigating them
632
00:22:34,840 –> 00:22:35,880
every week.
633
00:22:35,880 –> 00:22:37,800
Revenue means this, margin means that.
634
00:22:37,800 –> 00:22:39,600
SLA means exactly this time window
635
00:22:39,600 –> 00:22:40,960
with exactly these exclusions.
636
00:22:40,960 –> 00:22:41,800
And we’re done.
637
00:22:41,800 –> 00:22:43,200
Without that layer, one page KPI
638
00:22:43,200 –> 00:22:44,960
is becomes nine different calculations
639
00:22:44,960 –> 00:22:46,080
that happen to share a label.
640
00:22:46,080 –> 00:22:48,160
The visuals line up, the meanings don’t,
641
00:22:48,160 –> 00:22:50,640
and the minute you attach automation to it,
642
00:22:50,640 –> 00:22:52,160
you’ve automated disagreement.
643
00:22:52,160 –> 00:22:53,680
Layer three is state.
644
00:22:53,680 –> 00:22:55,280
This is the part dashboards never have,
645
00:22:55,280 –> 00:22:57,480
which is why dashboards can’t be control systems.
646
00:22:57,480 –> 00:22:59,520
State is operational memory.
647
00:22:59,520 –> 00:23:01,720
A report can tell you the variance is 9%.
648
00:23:01,720 –> 00:23:03,600
It cannot tell you whether the variance trigger
649
00:23:03,600 –> 00:23:06,640
already fired yesterday, whether an action was assigned,
650
00:23:06,640 –> 00:23:09,440
whether it’s in progress, whether it was overridden,
651
00:23:09,440 –> 00:23:13,280
or whether it failed silently in someone’s inbox.
652
00:23:13,280 –> 00:23:14,920
A decision engine needs a ledger.
653
00:23:14,920 –> 00:23:17,840
Trigger fired, owner assigned, action launched,
654
00:23:17,840 –> 00:23:19,600
status tracked, outcome recorded,
655
00:23:19,600 –> 00:23:20,840
exception reason captured.
656
00:23:20,840 –> 00:23:23,320
If you don’t have that, you don’t have determinism.
657
00:23:23,320 –> 00:23:25,000
You have a screenshot and a prayer.
658
00:23:25,000 –> 00:23:26,720
State also gives you audit posture.
659
00:23:26,720 –> 00:23:28,520
Not we think we did something.
660
00:23:28,520 –> 00:23:30,640
Actual traceability, who changed what,
661
00:23:30,640 –> 00:23:32,320
when and what the system did as a result.
662
00:23:32,320 –> 00:23:33,800
That’s the difference between governance
663
00:23:33,800 –> 00:23:35,280
and a PowerPoint claim of governance.
664
00:23:35,280 –> 00:23:36,800
Layer four is action.
665
00:23:36,800 –> 00:23:39,760
This is where intent stops being a meeting
666
00:23:39,760 –> 00:23:41,320
and becomes enforcement.
667
00:23:41,320 –> 00:23:43,040
Action is not send an email.
668
00:23:43,040 –> 00:23:45,680
Action is execute the pre-committed pathway,
669
00:23:45,680 –> 00:23:48,360
escalate to the accountable role, create the work item,
670
00:23:48,360 –> 00:23:51,560
apply the spend guard rail, trigger the remediation playbook,
671
00:23:51,560 –> 00:23:54,800
root approvals only where risk demands human judgment,
672
00:23:54,800 –> 00:23:55,960
and log everything.
673
00:23:55,960 –> 00:23:57,800
And yes, this is where organizations panic
674
00:23:57,800 –> 00:23:59,760
because action removes the ability to delay.
675
00:23:59,760 –> 00:24:01,960
It removes the ability to get alignment
676
00:24:01,960 –> 00:24:03,680
as a substitute for doing the work.
677
00:24:03,680 –> 00:24:04,680
But that’s the point.
678
00:24:04,680 –> 00:24:06,560
The system should only escalate to humans
679
00:24:06,560 –> 00:24:08,120
when the system cannot decide.
680
00:24:08,120 –> 00:24:09,400
Everything else is automation
681
00:24:09,400 –> 00:24:11,400
because everything else is repeatable.
682
00:24:11,400 –> 00:24:13,400
Layer five is interface.
683
00:24:13,400 –> 00:24:16,120
And this is where the one page KPI fantasy belongs.
684
00:24:16,120 –> 00:24:17,400
At the end, not the beginning,
685
00:24:17,400 –> 00:24:20,360
the interface is how executives access the decision engine.
686
00:24:20,360 –> 00:24:21,640
Sometimes that’s a dashboard.
687
00:24:21,640 –> 00:24:23,360
Sometimes it’s a scorecard.
688
00:24:23,360 –> 00:24:25,160
Increasingly it’s conversational.
689
00:24:25,160 –> 00:24:27,000
Are we within our risk tolerance?
690
00:24:27,000 –> 00:24:28,600
Which triggers fired this week?
691
00:24:28,600 –> 00:24:30,000
What actions are overdue?
692
00:24:30,000 –> 00:24:31,240
And who owns them?
693
00:24:31,240 –> 00:24:32,680
Here’s the key constraint.
694
00:24:32,680 –> 00:24:35,560
The interface must query decisions, not data sets.
695
00:24:35,560 –> 00:24:38,280
If the interface lets leaders wander through raw data,
696
00:24:38,280 –> 00:24:40,720
you just rebuild the debate loop in a nicer wrapper.
697
00:24:40,720 –> 00:24:43,040
If the interface surfaces decision state
698
00:24:43,040 –> 00:24:45,840
what the system concluded, what it did, what’s pending,
699
00:24:45,840 –> 00:24:47,600
you finally given them the control plane
700
00:24:47,600 –> 00:24:50,360
they were asking for badly when they said one page.
701
00:24:50,360 –> 00:24:52,200
So the stack is the deliverable.
702
00:24:52,200 –> 00:24:54,800
Data convergence eliminates contradictory inputs.
703
00:24:54,800 –> 00:24:57,560
Logic eliminates contradictory definitions.
704
00:24:57,560 –> 00:24:59,480
State eliminates contradictory narratives
705
00:24:59,480 –> 00:25:00,400
about what happened.
706
00:25:00,400 –> 00:25:02,800
Action eliminates contradictory response patterns.
707
00:25:02,800 –> 00:25:05,480
Interface eliminates contradictory access parts.
708
00:25:05,480 –> 00:25:07,160
And when you build it in that order,
709
00:25:07,160 –> 00:25:09,680
the executive experience becomes boring in the best way.
710
00:25:09,680 –> 00:25:12,400
Same condition, same outcome, same accountability.
711
00:25:12,400 –> 00:25:15,200
No interpretive freedom where the business can’t afford it.
712
00:25:15,200 –> 00:25:17,920
Now we can map these layers onto Microsoft components
713
00:25:17,920 –> 00:25:20,360
without turning this into a tutorial.
714
00:25:20,360 –> 00:25:23,600
Layer one, data convergence, fabric.
715
00:25:23,600 –> 00:25:25,560
One lake is entropy containment.
716
00:25:25,560 –> 00:25:28,680
Layer one is where most organizations accidentally sabotage
717
00:25:28,680 –> 00:25:32,080
determinism before they even get to the argument about KPIs.
718
00:25:32,080 –> 00:25:34,840
They treat data storage as an implementation detail.
719
00:25:34,840 –> 00:25:36,400
It is not.
720
00:25:36,400 –> 00:25:39,440
Data convergence is a contract where the organization agrees
721
00:25:39,440 –> 00:25:41,640
truth is allowed to live, how it’s curated,
722
00:25:41,640 –> 00:25:43,800
and under what governance it can graduate from
723
00:25:43,800 –> 00:25:45,960
interesting to decision grade.
724
00:25:45,960 –> 00:25:48,320
This is why fabric and one lake matter in this episode.
725
00:25:48,320 –> 00:25:49,720
Not because they’re trendy
726
00:25:49,720 –> 00:25:51,320
and not because the UI looks nicer
727
00:25:51,320 –> 00:25:53,200
than the last platform you regret buying.
728
00:25:53,200 –> 00:25:55,440
They matter because one lake is the closest thing
729
00:25:55,440 –> 00:25:58,120
Microsoft gives you to a single logical place
730
00:25:58,120 –> 00:26:00,400
where cross-domain metrics can be assembled
731
00:26:00,400 –> 00:26:03,320
without inventing a new copy of reality for every team.
732
00:26:03,320 –> 00:26:06,080
And yes, single logical place is doing a lot of work
733
00:26:06,080 –> 00:26:07,920
in that sentence because the failure mode
734
00:26:07,920 –> 00:26:09,560
isn’t that data doesn’t exist.
735
00:26:09,560 –> 00:26:12,680
The failure mode is that the same fact exists three times.
736
00:26:12,680 –> 00:26:15,920
Once in the source system, once in somebody’s export,
737
00:26:15,920 –> 00:26:18,160
and once in the clean data set
738
00:26:18,160 –> 00:26:20,000
that only one analyst understands.
739
00:26:20,000 –> 00:26:22,400
Each copy drifts, each copy refreshes differently,
740
00:26:22,400 –> 00:26:24,760
each copy produces a slightly different KPI,
741
00:26:24,760 –> 00:26:26,640
and then your executive meeting becomes
742
00:26:26,640 –> 00:26:29,360
a reconciliation workshop disguised as strategy
743
00:26:29,360 –> 00:26:30,400
that is entropy.
744
00:26:30,400 –> 00:26:33,680
Fabric gives you a way to contain it,
745
00:26:33,680 –> 00:26:36,320
but only if you treat it as architecture, not plumbing.
746
00:26:36,320 –> 00:26:39,160
The moment you allow every domain to build its own lake,
747
00:26:39,160 –> 00:26:40,800
you didn’t implement convergence,
748
00:26:40,800 –> 00:26:43,800
you implemented federated confusion with better branding.
749
00:26:43,800 –> 00:26:46,720
So the real move is this, establish one lake
750
00:26:46,720 –> 00:26:48,600
as the where truth lives boundary,
751
00:26:48,600 –> 00:26:49,720
not forever, not for everything,
752
00:26:49,720 –> 00:26:52,600
but for anything you intend to use as a deterministic trigger.
753
00:26:52,600 –> 00:26:53,920
If a metric can wake someone up,
754
00:26:53,920 –> 00:26:55,720
freeze, spend, or escalate an incident,
755
00:26:55,720 –> 00:26:58,040
it does not get to come from a best effort pipeline.
756
00:26:58,040 –> 00:26:59,640
It does not get to come from a workbook.
757
00:26:59,640 –> 00:27:02,680
It does not get to come from, “We’ll fix it next sprint.”
758
00:27:02,680 –> 00:27:05,160
Decision-grade inputs need an explicit life cycle.
759
00:27:05,160 –> 00:27:07,880
That’s where certified data products enter the conversation,
760
00:27:07,880 –> 00:27:11,000
not as governance theater, as anti-entropy mechanisms.
761
00:27:11,000 –> 00:27:14,120
The point of a certified product
762
00:27:14,120 –> 00:27:17,200
is to give the organization a trusted interface.
763
00:27:17,200 –> 00:27:20,680
This data set has an owner, a refresh expectation,
764
00:27:20,680 –> 00:27:23,480
documented meaning, and lineage you can defend
765
00:27:23,480 –> 00:27:25,720
when someone inevitably challenges it.
766
00:27:25,720 –> 00:27:27,760
And without that, your decision engine becomes
767
00:27:27,760 –> 00:27:29,120
probabilistic at the first layer.
768
00:27:29,120 –> 00:27:31,440
You end up automating actions of a data set
769
00:27:31,440 –> 00:27:33,680
that silently changed or refreshed late
770
00:27:33,680 –> 00:27:35,200
or started excluding a segment
771
00:27:35,200 –> 00:27:37,240
because someone tweaked a filter upstream.
772
00:27:37,240 –> 00:27:40,200
The system will still behave, it will just behave randomly.
773
00:27:40,200 –> 00:27:42,680
Now there’s a specific pathology fabric helps you expose
774
00:27:42,680 –> 00:27:44,240
in consistent refresh cadence.
775
00:27:44,240 –> 00:27:47,080
Executives think refresh cadence is a performance detail.
776
00:27:47,080 –> 00:27:50,080
Architects no refresh cadence is a decision integrity detail.
777
00:27:50,080 –> 00:27:53,200
If one KPI updates hourly and the other updates daily,
778
00:27:53,200 –> 00:27:55,920
your one page overview is a collage of different timelines
779
00:27:55,920 –> 00:27:57,480
pretending to be one moment in time.
780
00:27:57,480 –> 00:27:58,920
That’s not a dashboard problem.
781
00:27:58,920 –> 00:28:00,240
That’s a time model problem.
782
00:28:00,240 –> 00:28:02,520
And it turns deterministic triggers into traps.
783
00:28:02,520 –> 00:28:05,280
You think you’re detecting forecast variants today,
784
00:28:05,280 –> 00:28:07,440
but half the inputs are from yesterday.
785
00:28:07,440 –> 00:28:09,200
You think you’re preventing an SLA breach,
786
00:28:09,200 –> 00:28:11,920
but the incident state you’re looking at is stale.
787
00:28:11,920 –> 00:28:14,520
So in this layer, freshness is not a nice to have.
788
00:28:14,520 –> 00:28:16,280
It is a constraint you have to declare
789
00:28:16,280 –> 00:28:17,600
and force and monitor.
790
00:28:17,600 –> 00:28:19,480
If you can’t meet the freshness constraint,
791
00:28:19,480 –> 00:28:21,600
the KPI is not eligible for automation.
792
00:28:21,600 –> 00:28:22,640
Hush, yes.
793
00:28:22,640 –> 00:28:23,560
Also correct.
794
00:28:23,560 –> 00:28:26,200
Per view comes in here, but not as a compliance lecture.
795
00:28:26,200 –> 00:28:28,880
Per view is how you stop pretending lineage is optional.
796
00:28:28,880 –> 00:28:31,480
It’s how you attach traceability to the data product
797
00:28:31,480 –> 00:28:33,560
so your future self can answer the only question
798
00:28:33,560 –> 00:28:35,920
auditors and incident reviewers ever ask.
799
00:28:35,920 –> 00:28:38,280
Where did this number come from and who is responsible
800
00:28:38,280 –> 00:28:39,080
for its meaning?
801
00:28:39,080 –> 00:28:40,600
Because if you can’t answer that,
802
00:28:40,600 –> 00:28:42,160
you do not have decision architecture.
803
00:28:42,160 –> 00:28:43,560
You have curated screenshots.
804
00:28:43,560 –> 00:28:46,200
So the deliverable for layer one isn’t a lake house.
805
00:28:46,200 –> 00:28:49,440
It’s a convergence boundary with explicit eligibility rules,
806
00:28:49,440 –> 00:28:52,160
certified data sets, known refresh contracts,
807
00:28:52,160 –> 00:28:54,000
documented ownership, and lineage
808
00:28:54,000 –> 00:28:55,960
that survives personnel changes.
809
00:28:55,960 –> 00:28:59,360
Once you have that, the KPI stops being a negotiated artifact
810
00:28:59,360 –> 00:29:02,200
and starts being an input the system can legally trust.
811
00:29:02,200 –> 00:29:04,240
And now we hit the next inevitability.
812
00:29:04,240 –> 00:29:06,960
Even perfect convergence fails if the meaning layer drifts.
813
00:29:06,960 –> 00:29:10,560
So the next layer is the semantic model, the meaning compiler.
814
00:29:10,560 –> 00:29:13,640
Layer two, logic, power, BI semantic model,
815
00:29:13,640 –> 00:29:15,800
as the meaning compiler, layer two
816
00:29:15,800 –> 00:29:18,800
is logic and this is where most KPI programs quietly die
817
00:29:18,800 –> 00:29:20,920
not because the math is hard because the organization
818
00:29:20,920 –> 00:29:23,800
refuses to pick one definition and make it hurt.
819
00:29:23,800 –> 00:29:25,880
People talk about the data set like it’s the truth.
820
00:29:25,880 –> 00:29:28,080
It isn’t, raw data is just facts with sharp edges.
821
00:29:28,080 –> 00:29:31,040
Logic is where you sand those edges down into business meaning,
822
00:29:31,040 –> 00:29:34,960
revenue, margin, SLA, churn, forecast accuracy.
823
00:29:34,960 –> 00:29:36,120
Those aren’t fields.
824
00:29:36,120 –> 00:29:37,640
They’re agreements.
825
00:29:37,640 –> 00:29:40,120
And agreements are what entropy destroys first.
826
00:29:40,120 –> 00:29:42,200
In this stack, the power BI semantic model
827
00:29:42,200 –> 00:29:45,240
is the meaning compiler, not a model in the diagram sense.
828
00:29:45,240 –> 00:29:46,720
A compiler.
829
00:29:46,720 –> 00:29:50,320
It takes messy reality and outputs governed reusable definitions
830
00:29:50,320 –> 00:29:51,880
that the organization can’t casually
831
00:29:51,880 –> 00:29:53,840
reinterpret based on who’s in the meeting.
832
00:29:53,840 –> 00:29:56,640
If the same term can mean five things, it will mean five things.
833
00:29:56,640 –> 00:29:59,520
And then your KPI one-pager becomes a distributed argument
834
00:29:59,520 –> 00:30:00,120
surface.
835
00:30:00,120 –> 00:30:02,880
So the job of the semantic model is brutally simple.
836
00:30:02,880 –> 00:30:04,920
One definition of the thing that matters.
837
00:30:04,920 –> 00:30:07,880
One version named owned reused.
838
00:30:07,880 –> 00:30:10,760
One definition of revenue, one definition of forecast variance,
839
00:30:10,760 –> 00:30:14,640
one definition of SLA clock starts here, stops there.
840
00:30:14,640 –> 00:30:16,960
Because the moment you allow local flexibility,
841
00:30:16,960 –> 00:30:18,480
you create conditional chaos.
842
00:30:18,480 –> 00:30:19,840
The dashboard still renders.
843
00:30:19,840 –> 00:30:20,760
The color still flip.
844
00:30:20,760 –> 00:30:23,280
But the organization stops being deterministic.
845
00:30:23,280 –> 00:30:24,600
It becomes probabilistic.
846
00:30:24,600 –> 00:30:26,760
You get different answers depending on which team
847
00:30:26,760 –> 00:30:29,280
built the measure, which filter context they assumed,
848
00:30:29,280 –> 00:30:31,520
and which exceptions they helpfully embedded.
849
00:30:31,520 –> 00:30:34,000
And the thing most people miss is that self-service is not
850
00:30:34,000 –> 00:30:34,800
the enemy.
851
00:30:34,800 –> 00:30:36,480
Ungoverned self-service is.
852
00:30:36,480 –> 00:30:39,520
Self-service on top of a controlled semantic layer is fine.
853
00:30:39,520 –> 00:30:41,200
That’s just people asking questions.
854
00:30:41,200 –> 00:30:43,000
Self-service without semantic control is
855
00:30:43,000 –> 00:30:46,160
how you end up with 12 revenue numbers that all reconcile
856
00:30:46,160 –> 00:30:49,240
once you understand the logic, which is another way of saying
857
00:30:49,240 –> 00:30:51,000
nobody actually agrees.
858
00:30:51,000 –> 00:30:53,640
This clicked the first time someone tried to automate an action
859
00:30:53,640 –> 00:30:54,880
of a KPI.
860
00:30:54,880 –> 00:30:56,920
And it failed in the most predictable way.
861
00:30:56,920 –> 00:30:58,680
The KPI meant one thing in the report
862
00:30:58,680 –> 00:31:01,440
and a slightly different thing in the flow that consumed it.
863
00:31:01,440 –> 00:31:02,800
Nobody did anything malicious.
864
00:31:02,800 –> 00:31:05,040
The system just did what distributed systems do.
865
00:31:05,040 –> 00:31:06,320
It amplified ambiguity.
866
00:31:06,320 –> 00:31:09,640
So treat measures as policy, not calculations.
867
00:31:09,640 –> 00:31:11,740
A KPI measure isn’t whatever finance
868
00:31:11,740 –> 00:31:13,040
prefers this quarter.
869
00:31:13,040 –> 00:31:16,040
It’s the encoded intent, inclusion rules, exclusion rules,
870
00:31:16,040 –> 00:31:18,220
time intelligence, currency normalization,
871
00:31:18,220 –> 00:31:19,900
treatment of credits and returns,
872
00:31:19,900 –> 00:31:22,040
and whatever other compromises your business has been
873
00:31:22,040 –> 00:31:23,920
pretending aren’t compromises.
874
00:31:23,920 –> 00:31:25,920
Once the measure exists, it becomes a contract.
875
00:31:25,920 –> 00:31:27,580
And contracts need change control.
876
00:31:27,580 –> 00:31:29,580
This is where governance stops being a document
877
00:31:29,580 –> 00:31:31,380
and becomes a design constraint.
878
00:31:31,380 –> 00:31:33,100
Your semantic model needs a lifecycle,
879
00:31:33,100 –> 00:31:34,820
draft review, approve, publish.
880
00:31:34,820 –> 00:31:37,280
And when it changes, it changes intentionally
881
00:31:37,280 –> 00:31:39,220
with a reason you can explain six months later.
882
00:31:39,220 –> 00:31:41,260
Because if you can’t explain the change,
883
00:31:41,260 –> 00:31:43,380
you can’t defend the decisions that followed it.
884
00:31:43,380 –> 00:31:45,900
Now there’s a practical boundary in the Microsoft ecosystem
885
00:31:45,900 –> 00:31:47,780
that lines up perfectly with this.
886
00:31:47,780 –> 00:31:51,140
The approved for co-pilot concept on semantic models.
887
00:31:51,140 –> 00:31:53,140
Whatever you call the flag in your tenant,
888
00:31:53,140 –> 00:31:54,860
the architectural meaning is the same.
889
00:31:54,860 –> 00:31:57,220
You’re declaring a trust boundary for reuse.
890
00:31:57,220 –> 00:31:58,780
Not everything should be reusable.
891
00:31:58,780 –> 00:32:00,220
Not everything should be queryable
892
00:32:00,220 –> 00:32:01,780
by an executive facing agent.
893
00:32:01,780 –> 00:32:03,300
Not every half finished measure
894
00:32:03,300 –> 00:32:05,940
should become the basis for automated escalation.
895
00:32:05,940 –> 00:32:07,620
The approved model becomes the one
896
00:32:07,620 –> 00:32:10,220
the organization treats as decision grade logic.
897
00:32:10,220 –> 00:32:11,940
It’s the place where the compiler output
898
00:32:11,940 –> 00:32:15,460
is stable enough to feed the next layers, state and action.
899
00:32:15,460 –> 00:32:17,780
And this is also why you don’t want your decision engine
900
00:32:17,780 –> 00:32:21,300
glued together with ad hoc DAX scattered across 50 reports.
901
00:32:21,300 –> 00:32:22,820
That architecture guarantees drift.
902
00:32:22,820 –> 00:32:25,180
Reports get copied, measures get tweaked.
903
00:32:25,180 –> 00:32:27,500
Somebody fixes just this page.
904
00:32:27,500 –> 00:32:28,660
And now you have logic folks.
905
00:32:28,660 –> 00:32:31,580
Forks are entropy generators, centralized the definitions,
906
00:32:31,580 –> 00:32:34,420
reuse them and make exceptions explicit.
907
00:32:34,420 –> 00:32:37,900
Because exception in the measure is how ambiguity hides.
908
00:32:37,900 –> 00:32:39,940
If you need a special case rule for a segment,
909
00:32:39,940 –> 00:32:41,420
you document it, you version it,
910
00:32:41,420 –> 00:32:43,140
and you treat it as a business decision.
911
00:32:43,140 –> 00:32:46,340
Not as a silent adjustment that only one analyst remembers.
912
00:32:46,340 –> 00:32:47,900
Now here’s the uncomfortable part.
913
00:32:47,900 –> 00:32:49,780
Semantics are where politics show up.
914
00:32:49,780 –> 00:32:51,900
People will fight to keep interpretive freedom
915
00:32:51,900 –> 00:32:54,340
because interpretive freedom is how they manage their own risk.
916
00:32:54,340 –> 00:32:57,020
If the KPI is deterministic, excuses evaporate.
917
00:32:57,020 –> 00:33:00,100
If the KPI is deterministic, ownership becomes uncomfortable.
918
00:33:00,100 –> 00:33:01,660
If the KPI is deterministic,
919
00:33:01,660 –> 00:33:04,540
the system can enforce consequences without asking permission.
920
00:33:04,540 –> 00:33:06,300
That’s why semantic drift is inevitable
921
00:33:06,300 –> 00:33:07,540
unless you design against it.
922
00:33:07,540 –> 00:33:11,340
So you enforce semantic determinism the same way you enforce security determinism.
923
00:33:11,340 –> 00:33:15,660
Single source, least privilege, review gates, and auditability.
924
00:33:15,660 –> 00:33:17,660
You don’t encourage consistency.
925
00:33:17,660 –> 00:33:19,500
You make inconsistency expensive.
926
00:33:19,500 –> 00:33:21,980
And once you do, you unlock the real payoff.
927
00:33:21,980 –> 00:33:24,300
The meeting stops being about whose number is correct
928
00:33:24,300 –> 00:33:26,580
and starts being about what the system will do next
929
00:33:26,580 –> 00:33:28,140
when the number crosses a line.
930
00:33:28,140 –> 00:33:29,460
But there’s still a missing piece.
931
00:33:29,460 –> 00:33:32,420
Even perfectly governed meaning still doesn’t create a decision engine
932
00:33:32,420 –> 00:33:34,340
because decisions require memory.
933
00:33:34,340 –> 00:33:37,500
Reports forget dashboards forget conversations forget.
934
00:33:37,500 –> 00:33:39,220
So the next layer is state.
935
00:33:39,220 –> 00:33:42,460
An operational ledger that remembers what fired, what happened,
936
00:33:42,460 –> 00:33:44,700
and whether anyone actually did the work.
937
00:33:44,700 –> 00:33:48,380
Layer three, state, data verse as operational decision ledger.
938
00:33:48,380 –> 00:33:52,300
Layer three is state and it’s the part executives are unknowingly begging for
939
00:33:52,300 –> 00:33:54,340
when they ask for KPI’s on one page
940
00:33:54,340 –> 00:33:57,300
because they don’t actually want to stare at yesterday’s numbers.
941
00:33:57,300 –> 00:33:59,700
They want to know what the organization is doing about them.
942
00:33:59,700 –> 00:34:01,180
A report can show a variance.
943
00:34:01,180 –> 00:34:02,420
It can’t show obligation.
944
00:34:02,420 –> 00:34:05,180
It can’t show whether anyone accepted responsibility,
945
00:34:05,180 –> 00:34:06,860
whether the response is underway
946
00:34:06,860 –> 00:34:10,580
or whether the will look into it quietly died in someone’s inbox.
947
00:34:10,580 –> 00:34:12,580
Dashboards don’t retain operational memory.
948
00:34:12,580 –> 00:34:15,140
They refresh, they overwrite, they forget.
949
00:34:15,140 –> 00:34:18,340
And the organization uses that amnesia as a coping mechanism.
950
00:34:18,340 –> 00:34:20,940
If there’s no durable record of the trigger fired
951
00:34:20,940 –> 00:34:24,140
and you were accountable, then accountability becomes vibes.
952
00:34:24,140 –> 00:34:26,780
It becomes whatever story gets told in the next meeting.
953
00:34:26,780 –> 00:34:28,660
The system becomes politically editable,
954
00:34:28,660 –> 00:34:30,580
which is another way of saying probabilistic.
955
00:34:30,580 –> 00:34:32,060
A decision engine can’t afford that.
956
00:34:32,060 –> 00:34:34,900
State is what turns a decision rule into a trackable event
957
00:34:34,900 –> 00:34:38,540
with consequences. It’s the difference between we noticed and we acted.
958
00:34:38,540 –> 00:34:41,140
And it’s also the difference between we think we’re governed
959
00:34:41,140 –> 00:34:42,700
and we can prove we are.
960
00:34:42,700 –> 00:34:44,380
In the Microsoft ecosystem,
961
00:34:44,380 –> 00:34:46,660
Dataverse is the cleanest place to hold this,
962
00:34:46,660 –> 00:34:47,740
not because it’s magical,
963
00:34:47,740 –> 00:34:51,020
but because it’s designed to be an operational system of record
964
00:34:51,020 –> 00:34:55,140
with security, auditing and a schema you can actually defend.
965
00:34:55,140 –> 00:34:56,700
And that distinction matters.
966
00:34:56,700 –> 00:34:59,620
Fabric and Power BI handle analytic state,
967
00:34:59,620 –> 00:35:02,180
facts, aggregates, trends, historical slices.
968
00:35:02,180 –> 00:35:04,940
That’s valuable, but it’s not operational state.
969
00:35:04,940 –> 00:35:06,980
Operational state is about commitments.
970
00:35:06,980 –> 00:35:09,180
What decision was made, what action was created
971
00:35:09,180 –> 00:35:11,180
and what the current status is right now.
972
00:35:11,180 –> 00:35:13,020
So what goes into the decision ledger?
973
00:35:13,020 –> 00:35:14,460
At minimum, the trigger instance,
974
00:35:14,460 –> 00:35:16,060
the rule version that evaluated it,
975
00:35:16,060 –> 00:35:18,620
the time stamp it fired and the context it fired under,
976
00:35:18,620 –> 00:35:21,980
the segment, the service, the region, the severity,
977
00:35:21,980 –> 00:35:24,100
whatever constraints made it eligible.
978
00:35:24,100 –> 00:35:26,540
Then the owner assignment, not a team,
979
00:35:26,540 –> 00:35:28,780
not a channel, not someone in finance,
980
00:35:28,780 –> 00:35:31,900
a role mapped to an actual accountable identity at runtime.
981
00:35:31,900 –> 00:35:34,420
And then the action, which action pathway was selected,
982
00:35:34,420 –> 00:35:36,940
which playbook, which guardrail, which workflow,
983
00:35:36,940 –> 00:35:38,660
and whether it executed automatically
984
00:35:38,660 –> 00:35:41,540
or required a human override, then the life cycle,
985
00:35:41,540 –> 00:35:44,780
status changes in progress, waiting on approval, escalated,
986
00:35:44,780 –> 00:35:47,740
completed, failed, overwritten, expired.
987
00:35:47,740 –> 00:35:50,380
And then the outcome, the post-action measurement,
988
00:35:50,380 –> 00:35:52,660
did the variance stabilize inside tolerance,
989
00:35:52,660 –> 00:35:55,340
did SLA breach risk drop, did the system recover,
990
00:35:55,340 –> 00:35:56,900
did the action reduce forecast error
991
00:35:56,900 –> 00:35:58,420
or just generate activity.
992
00:35:58,420 –> 00:36:00,700
Without that last part, you built an alarm system,
993
00:36:00,700 –> 00:36:01,820
not a control system.
994
00:36:01,820 –> 00:36:04,100
Now here’s where most organizations mess up.
995
00:36:04,100 –> 00:36:06,820
They store this state in the same place as they store
996
00:36:06,820 –> 00:36:09,540
everything else, email, chat, meeting notes,
997
00:36:09,540 –> 00:36:12,100
tickets, random spreadsheets, and then they act surprised
998
00:36:12,100 –> 00:36:14,340
when nobody can reconstruct what happened during an audit
999
00:36:14,340 –> 00:36:15,420
or an incident review.
1000
00:36:15,420 –> 00:36:17,620
Those tools are communication tools.
1001
00:36:17,620 –> 00:36:18,620
They are not ledgers.
1002
00:36:18,620 –> 00:36:20,340
A ledger is a place where state changes
1003
00:36:20,340 –> 00:36:22,260
are explicit, durable and queryable.
1004
00:36:22,260 –> 00:36:25,020
It’s where we did something, becomes here is the record.
1005
00:36:25,020 –> 00:36:26,660
Dataverse gives you that model.
1006
00:36:26,660 –> 00:36:29,140
Tables for decision events, assignments, actions,
1007
00:36:29,140 –> 00:36:29,980
and outcomes.
1008
00:36:29,980 –> 00:36:33,460
Relationships between them, security roles, change tracking.
1009
00:36:33,460 –> 00:36:36,100
And most importantly, a way to make the decision system
1010
00:36:36,100 –> 00:36:39,180
observable without forcing executives to become archaeologists.
1011
00:36:39,180 –> 00:36:41,460
Because executives don’t want raw telemetry.
1012
00:36:41,460 –> 00:36:42,460
They want posture.
1013
00:36:42,460 –> 00:36:44,780
They want to ask, which triggers fired this week?
1014
00:36:44,780 –> 00:36:48,180
Which actions are overdue and which ones were overwritten?
1015
00:36:48,180 –> 00:36:50,900
And they want a coherent answer that doesn’t require a meeting
1016
00:36:50,900 –> 00:36:51,660
to assemble.
1017
00:36:51,660 –> 00:36:54,780
A decision ledger also fixes another quiet failure mode.
1018
00:36:54,780 –> 00:36:56,180
Semantic drift over time.
1019
00:36:56,180 –> 00:36:58,860
If you don’t store which rule version fired the trigger,
1020
00:36:58,860 –> 00:37:01,660
you can’t explain past decisions after the logic changes.
1021
00:37:01,660 –> 00:37:04,180
Your future self will look at an old action and ask,
1022
00:37:04,180 –> 00:37:06,140
why did we freeze spent here?
1023
00:37:06,140 –> 00:37:08,220
And you won’t know because the KPI definition
1024
00:37:08,220 –> 00:37:09,940
has changed three times since then.
1025
00:37:09,940 –> 00:37:12,140
That is how systems become legally indefensible.
1026
00:37:12,140 –> 00:37:14,700
So the ledger has to store the decision context
1027
00:37:14,700 –> 00:37:16,060
as it existed at the time.
1028
00:37:16,060 –> 00:37:19,460
The evaluated value, the threshold, the duration condition
1029
00:37:19,460 –> 00:37:20,580
and the rule version.
1030
00:37:20,580 –> 00:37:22,340
That’s how you make the system auditable.
1031
00:37:22,340 –> 00:37:23,820
Not by writing a policy document,
1032
00:37:23,820 –> 00:37:26,340
by capturing evidence as a byproduct of execution.
1033
00:37:26,340 –> 00:37:28,420
And it’s not just audits, it’s operations.
1034
00:37:29,380 –> 00:37:31,140
In the revenue variance scenario,
1035
00:37:31,140 –> 00:37:33,780
this ledger is how you prevent the same argument every month.
1036
00:37:33,780 –> 00:37:36,900
You can see, trigger fired on this date, owner assigned,
1037
00:37:36,900 –> 00:37:39,980
spend guardrail executed, pipeline acceleration started,
1038
00:37:39,980 –> 00:37:43,700
stabilization achieved or not, no mythology, just state.
1039
00:37:43,700 –> 00:37:46,100
In the IT/SLA scenario, it’s the same.
1040
00:37:46,100 –> 00:37:48,860
Breach countdown started, escalation tier triggered,
1041
00:37:48,860 –> 00:37:52,220
playbook launched, communication sent, resolution achieved,
1042
00:37:52,220 –> 00:37:55,060
and whether the system prevented breach or arrived late
1043
00:37:55,060 –> 00:37:56,620
and pretended it was fine.
1044
00:37:56,620 –> 00:37:58,100
This is the uncomfortable truth.
1045
00:37:58,100 –> 00:38:01,140
Without a decision ledger, you don’t have a decision engine.
1046
00:38:01,140 –> 00:38:03,100
You have a report that points at problems
1047
00:38:03,100 –> 00:38:05,500
and a human network that may or may not respond.
1048
00:38:05,500 –> 00:38:07,500
With a ledger, you now have a control surface
1049
00:38:07,500 –> 00:38:08,860
you can actually manage.
1050
00:38:08,860 –> 00:38:11,340
Backlog of triggered obligations, aging of actions,
1051
00:38:11,340 –> 00:38:14,700
override rates, false trigger rates, and intervention efficacy.
1052
00:38:14,700 –> 00:38:17,300
And once you have state, the next layer becomes inevitable.
1053
00:38:17,300 –> 00:38:18,620
Now you can execute.
1054
00:38:18,620 –> 00:38:19,700
Not notify.
1055
00:38:19,700 –> 00:38:21,860
Because there’s no point in storing obligations
1056
00:38:21,860 –> 00:38:23,500
if nothing ever enforces them.
1057
00:38:23,500 –> 00:38:24,300
Layer four.
1058
00:38:24,300 –> 00:38:24,900
Action.
1059
00:38:24,900 –> 00:38:26,940
Power automate as intent enforcement,
1060
00:38:26,940 –> 00:38:28,380
not convenience automation.
1061
00:38:28,380 –> 00:38:29,620
Layer four is action.
1062
00:38:29,620 –> 00:38:31,660
And this is where most KPI programs
1063
00:38:31,660 –> 00:38:33,820
quietly refuse to grow up.
1064
00:38:33,820 –> 00:38:37,100
They stop at alerts, they send emails, they post in teams,
1065
00:38:37,100 –> 00:38:38,580
they create a notification storm
1066
00:38:38,580 –> 00:38:40,860
and then call it operationalizing data.
1067
00:38:40,860 –> 00:38:43,820
That is an action that’s anxiety delivery.
1068
00:38:43,820 –> 00:38:46,820
Action means the system executes the pre-committed pathway.
1069
00:38:46,820 –> 00:38:49,940
On time, the same way, every time the trigger condition is met,
1070
00:38:49,940 –> 00:38:52,540
not when someone feels like it, not when the meeting happens,
1071
00:38:52,540 –> 00:38:55,460
not when the one person who understands the report is online.
1072
00:38:55,460 –> 00:38:57,060
In the Microsoft ecosystem,
1073
00:38:57,060 –> 00:38:59,380
power automate is the obvious execution layer,
1074
00:38:59,380 –> 00:39:01,500
but only if you treat it like enforcement,
1075
00:39:01,500 –> 00:39:03,180
not like personal productivity.
1076
00:39:03,180 –> 00:39:05,820
Because most organizations use power automate
1077
00:39:05,820 –> 00:39:07,020
as a convenience tool.
1078
00:39:07,020 –> 00:39:10,740
Route this form, copy this file, send this reminder.
1079
00:39:10,740 –> 00:39:15,020
Fine, harmless, also irrelevant to decision architecture.
1080
00:39:15,020 –> 00:39:18,140
In a decision engine, flows become policy actuators.
1081
00:39:18,140 –> 00:39:19,780
They take a deterministic trigger
1082
00:39:19,780 –> 00:39:22,260
and they enforce an obligation against an accountable role
1083
00:39:22,260 –> 00:39:24,620
with a time constraint and a recorded outcome.
1084
00:39:24,620 –> 00:39:27,220
So the first architectural decision is the trigger style.
1085
00:39:27,220 –> 00:39:29,820
You don’t trigger off vibes, you trigger off events.
1086
00:39:29,820 –> 00:39:31,900
Either the analytic layer detects a condition
1087
00:39:31,900 –> 00:39:33,340
and publishes an event,
1088
00:39:33,340 –> 00:39:36,420
or the operational layer detects a change and raises a signal.
1089
00:39:36,420 –> 00:39:37,580
But the point is the same.
1090
00:39:37,580 –> 00:39:39,260
The flow doesn’t check the dashboard.
1091
00:39:39,260 –> 00:39:41,220
It listens for decision-grade conditions.
1092
00:39:41,220 –> 00:39:43,580
And this is where the system either stays deterministic
1093
00:39:43,580 –> 00:39:45,300
or collapses into conditional chaos.
1094
00:39:45,300 –> 00:39:47,420
If you let every team build its own triggers,
1095
00:39:47,420 –> 00:39:49,660
you’ll get 20 variations of forecast variance
1096
00:39:49,660 –> 00:39:52,740
and 15 variations of SLA breach risk.
1097
00:39:52,740 –> 00:39:55,780
And your flow inventory will become another entropy generator.
1098
00:39:55,780 –> 00:39:57,380
So you centralize trigger definitions
1099
00:39:57,380 –> 00:39:59,420
the same way you centralize semantic logic,
1100
00:39:59,420 –> 00:40:02,700
versioned, owned, and tied back to the rule that generated them.
1101
00:40:02,700 –> 00:40:05,180
Once the trigger fires, the next problem is rooting.
1102
00:40:05,180 –> 00:40:06,900
Rooting is architecture.
1103
00:40:06,900 –> 00:40:11,140
Most companies treat routing as send it to a group.
1104
00:40:11,140 –> 00:40:14,100
That’s not routing, that’s abdication with an email address.
1105
00:40:14,100 –> 00:40:17,100
Deterministic routing means role-based assignment.
1106
00:40:17,100 –> 00:40:19,620
The trigger maps to an accountable role, that role maps
1107
00:40:19,620 –> 00:40:21,420
to a real person at runtime.
1108
00:40:21,420 –> 00:40:24,140
And the escalation path is defined before the incident,
1109
00:40:24,140 –> 00:40:26,260
before the forecast drift, before the meeting.
1110
00:40:26,260 –> 00:40:28,420
If the response depends on who noticed it first,
1111
00:40:28,420 –> 00:40:29,380
you don’t have a system.
1112
00:40:29,380 –> 00:40:30,100
You have luck.
1113
00:40:30,100 –> 00:40:32,260
So flows should do three things immediately
1114
00:40:32,260 –> 00:40:33,740
when a trigger arrives.
1115
00:40:33,740 –> 00:40:36,660
One, create or update the decision ledger record,
1116
00:40:36,660 –> 00:40:39,980
trigger-fired, rule-version timestamp context.
1117
00:40:39,980 –> 00:40:42,500
Two, assign ownership, one accountable role,
1118
00:40:42,500 –> 00:40:44,860
not a committee, not a police review.
1119
00:40:44,860 –> 00:40:45,780
Ownership.
1120
00:40:45,780 –> 00:40:47,620
Three, start the clock.
1121
00:40:47,620 –> 00:40:50,460
The time constraint becomes a timer, the system can enforce,
1122
00:40:50,460 –> 00:40:52,060
not a promise humans can ignore.
1123
00:40:52,060 –> 00:40:53,660
Now here’s where people get nervous.
1124
00:40:53,660 –> 00:40:56,220
Approvals, approvals are where determinism goes to die
1125
00:40:56,220 –> 00:40:57,980
because approvals invite debate loops.
1126
00:40:57,980 –> 00:41:00,340
But you still need humans in the system sometimes,
1127
00:41:00,340 –> 00:41:03,780
because some actions are genuinely high-risk, freezing spend,
1128
00:41:03,780 –> 00:41:07,300
customer communications, public incident declarations.
1129
00:41:07,300 –> 00:41:08,540
Those require judgment.
1130
00:41:08,540 –> 00:41:09,500
The trick is brutal.
1131
00:41:09,500 –> 00:41:11,740
Approvals only exist where risk demands them,
1132
00:41:11,740 –> 00:41:13,220
and they are bounded by time.
1133
00:41:13,220 –> 00:41:15,780
If your approval takes longer than the risk curve allows,
1134
00:41:15,780 –> 00:41:17,140
you didn’t add governance.
1135
00:41:17,140 –> 00:41:19,540
You added failure, so you use guardrails.
1136
00:41:19,540 –> 00:41:22,260
Auto-execute low-risk steps, reserve approval
1137
00:41:22,260 –> 00:41:23,780
for irreversible actions,
1138
00:41:23,780 –> 00:41:26,300
and escalate cleanly when the system hits ambiguity.
1139
00:41:26,300 –> 00:41:29,780
An escalation has to be a first-class outcome, not a failure state.
1140
00:41:29,780 –> 00:41:32,540
When the system cannot decide, it says so explicitly,
1141
00:41:32,540 –> 00:41:35,900
it records that it escalated, it records who it escalated to.
1142
00:41:35,900 –> 00:41:39,220
And it records whether the human accepted, overwrote, or delayed.
1143
00:41:39,220 –> 00:41:41,260
That’s how you keep the probabilistic parts
1144
00:41:41,260 –> 00:41:43,540
visible instead of letting them hide in chat.
1145
00:41:43,540 –> 00:41:45,460
Now tie it back to the two scenarios.
1146
00:41:45,460 –> 00:41:47,660
For revenue-focused variants, the trigger fires,
1147
00:41:47,660 –> 00:41:51,020
the ledger records it, ownership plans with the VP of RevOPS,
1148
00:41:51,020 –> 00:41:53,380
and the flow executes pre-committed actions
1149
00:41:53,380 –> 00:41:56,300
like creating a variance response work item,
1150
00:41:56,300 –> 00:41:57,980
applying a spend guardrail,
1151
00:41:57,980 –> 00:42:01,940
and launching a time-boxed pipeline acceleration playbook.
1152
00:42:01,940 –> 00:42:05,460
If it needs an exception, maybe a strategic campaign can’t be paused.
1153
00:42:05,460 –> 00:42:08,220
The flow routes an explicit exception approval with a deadline
1154
00:42:08,220 –> 00:42:09,220
and logs the reason.
1155
00:42:09,220 –> 00:42:11,140
For SLA breach prevention,
1156
00:42:11,140 –> 00:42:13,180
the trigger isn’t breach occurred.
1157
00:42:13,180 –> 00:42:14,140
That’s reporting.
1158
00:42:14,140 –> 00:42:17,340
The trigger is breach-risk threshold reached with a countdown.
1159
00:42:17,340 –> 00:42:19,620
The flow escalates by tear automatically,
1160
00:42:19,620 –> 00:42:22,300
opens the major incident bridge at the defined threshold,
1161
00:42:22,300 –> 00:42:25,380
pushes the playbook tasks, and updates state continuously.
1162
00:42:25,380 –> 00:42:27,900
No heroics, just clock-driven enforcement.
1163
00:42:27,900 –> 00:42:30,380
And yes, this is where the system becomes uncomfortable,
1164
00:42:30,380 –> 00:42:32,420
because it exposes the truth.
1165
00:42:32,420 –> 00:42:35,500
Half of what you call process is actually negotiation.
1166
00:42:35,500 –> 00:42:39,340
Automation removes negotiation by making the default response pre-committed.
1167
00:42:39,340 –> 00:42:40,580
People can still override.
1168
00:42:40,580 –> 00:42:42,740
They just have to do it in the open with a reason,
1169
00:42:42,740 –> 00:42:45,260
with a timestamp, against a rule version.
1170
00:42:45,260 –> 00:42:46,980
That’s governance that actually exists.
1171
00:42:46,980 –> 00:42:50,380
So power automate isn’t there to make work feel smoother.
1172
00:42:50,380 –> 00:42:52,380
It’s there to make intent enforceable.
1173
00:42:52,380 –> 00:42:53,540
Because without execution,
1174
00:42:53,540 –> 00:42:55,780
KPIs remain what they’ve always been.
1175
00:42:55,780 –> 00:42:58,060
Colored tiles that report failure politely.
1176
00:42:58,060 –> 00:43:02,100
With execution, the organization finally behaves like it has a control plane.
1177
00:43:02,100 –> 00:43:03,660
And now the last layer matters.
1178
00:43:03,660 –> 00:43:07,260
How leaders access all of this without becoming report archaeologists.
1179
00:43:07,260 –> 00:43:11,860
Layer 5, interface, copilot studio, as decision access, not report search.
1180
00:43:11,860 –> 00:43:17,420
Layer 5 is interface, and this is where most organizations commit the final fatal misunderstanding.
1181
00:43:17,420 –> 00:43:19,860
They think the interface is the product, it isn’t.
1182
00:43:19,860 –> 00:43:23,980
The interface is just how leadership touches the machine you built in layers one through four.
1183
00:43:23,980 –> 00:43:25,900
If the machine isn’t deterministic,
1184
00:43:25,900 –> 00:43:28,740
the interface can only beautify uncertainty.
1185
00:43:28,740 –> 00:43:30,380
And if the machine is deterministic,
1186
00:43:30,380 –> 00:43:31,900
the interface’s job is simple,
1187
00:43:31,900 –> 00:43:36,300
exposed decision posture without forcing executives to become part-time analysts.
1188
00:43:36,300 –> 00:43:37,980
This is why copilot studio matters here,
1189
00:43:37,980 –> 00:43:40,740
not as a chatbot, not as natural language for power BI.
1190
00:43:40,740 –> 00:43:44,340
And definitely not as an expensive way to ask what were sales last month.
1191
00:43:44,340 –> 00:43:47,540
Copilot studio becomes valuable when it access decision access.
1192
00:43:47,540 –> 00:43:50,180
Meaning executives query decisions and obligations,
1193
00:43:50,180 –> 00:43:51,580
not data sets and visuals.
1194
00:43:51,580 –> 00:43:54,460
Because the executive question is almost never what is the number.
1195
00:43:54,460 –> 00:43:55,700
That’s what analysts ask.
1196
00:43:55,700 –> 00:43:58,260
The executive question is, are we inside tolerance?
1197
00:43:58,260 –> 00:44:00,020
And if not, what is already in motion?
1198
00:44:00,020 –> 00:44:02,900
So the interface has to answer in the native language of leaders.
1199
00:44:02,900 –> 00:44:05,820
Risk, posture, exceptions, and accountability.
1200
00:44:05,820 –> 00:44:08,260
Not measures and slices.
1201
00:44:08,260 –> 00:44:10,300
If the decision engine is built correctly,
1202
00:44:10,300 –> 00:44:12,300
the interface can respond like a control plane.
1203
00:44:12,300 –> 00:44:14,180
It can say these triggers fired,
1204
00:44:14,180 –> 00:44:16,820
these owners were assigned, these actions are overdue,
1205
00:44:16,820 –> 00:44:18,180
these exceptions were granted,
1206
00:44:18,180 –> 00:44:20,100
and these interventions worked or didn’t.
1207
00:44:20,100 –> 00:44:23,300
And crucially, it can do that without inventing new logic.
1208
00:44:23,300 –> 00:44:26,900
This is where people get sloppy and accidentally reintroduce entropy.
1209
00:44:26,900 –> 00:44:30,100
They let the agent freestyle, they let it summarize the business.
1210
00:44:30,100 –> 00:44:32,460
They let it do what probabilistic systems do,
1211
00:44:32,460 –> 00:44:34,900
generate plausible text that feels authoritative.
1212
00:44:34,900 –> 00:44:38,180
That’s not decision access, that’s narrative generation.
1213
00:44:38,180 –> 00:44:40,020
So the architectural mandate is strict.
1214
00:44:40,020 –> 00:44:43,500
Copilot studio must be grounded in approved logic and state.
1215
00:44:43,500 –> 00:44:45,940
It should read from the semantic model for definitions
1216
00:44:45,940 –> 00:44:47,820
and from the decision ledger for what happened,
1217
00:44:47,820 –> 00:44:51,020
it should not become another place where meanings drift.
1218
00:44:51,020 –> 00:44:54,100
This is where the approved for copilot boundary is not a nice checkbox.
1219
00:44:54,100 –> 00:44:55,020
It’s a trust boundary.
1220
00:44:55,020 –> 00:44:57,620
If a model isn’t approved, the agent doesn’t use it.
1221
00:44:57,620 –> 00:45:00,420
If a metric definition isn’t governed, the agent doesn’t paraphrase it.
1222
00:45:00,420 –> 00:45:03,500
If an action isn’t in the ledger, the agent doesn’t claim it happened.
1223
00:45:03,500 –> 00:45:05,540
This is how you keep a conversational interface
1224
00:45:05,540 –> 00:45:07,460
from turning into conditional chaos.
1225
00:45:07,460 –> 00:45:10,380
Now, executives also don’t want one agent per system.
1226
00:45:10,380 –> 00:45:12,220
They don’t want to memorize which bot to ask.
1227
00:45:12,220 –> 00:45:15,420
They want a single entry point that roots to the right domain.
1228
00:45:15,420 –> 00:45:17,420
So you treat the interface as an orchestrator,
1229
00:45:17,420 –> 00:45:19,180
even if it looks like one chat window.
1230
00:45:19,180 –> 00:45:22,620
A lead agent handles the conversation and roots to domain agents.
1231
00:45:22,620 –> 00:45:26,060
Finance, operation, security, HR, whatever you can defend.
1232
00:45:26,060 –> 00:45:28,980
And each domain agent is constrained by its data products,
1233
00:45:28,980 –> 00:45:31,620
its semantic definitions and its permitted actions.
1234
00:45:31,620 –> 00:45:32,700
That last part matters.
1235
00:45:32,700 –> 00:45:34,980
Not what it can read, what it is allowed to do.
1236
00:45:34,980 –> 00:45:38,580
Because copilot everywhere without intent boundaries becomes unordered.
1237
00:45:38,580 –> 00:45:41,660
You end up with an assistant that can see everything, say anything,
1238
00:45:41,660 –> 00:45:44,620
and suggest actions that aren’t tied to pre-committed rules.
1239
00:45:44,620 –> 00:45:47,460
That’s not empowerment, that’s liability.
1240
00:45:47,460 –> 00:45:49,980
So the interface needs explicit guardrails,
1241
00:45:49,980 –> 00:45:53,460
allowed actions, required confirmations, and stop rules.
1242
00:45:53,460 –> 00:45:56,340
When confidence is low or the situation is out of model,
1243
00:45:56,340 –> 00:46:00,140
the agent escalates to a human and logs that escalation as an event.
1244
00:46:00,140 –> 00:46:02,140
It doesn’t guess, it doesn’t be helpful.
1245
00:46:02,140 –> 00:46:04,700
It stops.
1246
00:46:04,700 –> 00:46:08,020
That’s how the interface stays aligned with deterministic core
1247
00:46:08,020 –> 00:46:10,500
and uses probabilistic AI-weared belongs.
1248
00:46:10,500 –> 00:46:13,180
Explanation, summarization, option generation.
1249
00:46:13,180 –> 00:46:14,620
Never as the source of truth.
1250
00:46:14,620 –> 00:46:16,140
Never as the rule engine.
1251
00:46:16,140 –> 00:46:18,180
And once you do this, the executive experience
1252
00:46:18,180 –> 00:46:20,460
changes in a way dashboards never achieve.
1253
00:46:20,460 –> 00:46:23,220
Instead of open the report and interpret the interaction
1254
00:46:23,220 –> 00:46:27,140
becomes, are we within revenue risk tolerance for Q3?
1255
00:46:27,140 –> 00:46:30,420
Which variance triggers fired in the last 10 business days?
1256
00:46:30,420 –> 00:46:33,660
What actions did the system execute and what’s still waiting?
1257
00:46:33,660 –> 00:46:35,860
Show me exceptions and who approved them.
1258
00:46:35,860 –> 00:46:36,740
That’s not reporting.
1259
00:46:36,740 –> 00:46:38,620
That’s governance with a user interface.
1260
00:46:38,620 –> 00:46:40,180
And it collapses decision latency
1261
00:46:40,180 –> 00:46:42,980
because leaders no longer spend time finding the page,
1262
00:46:42,980 –> 00:46:45,660
interpreting the filters and arguing about the number.
1263
00:46:45,660 –> 00:46:48,340
They spend time on the only thing humans are still needed for,
1264
00:46:48,340 –> 00:46:50,700
choosing between explicit, bounded options
1265
00:46:50,700 –> 00:46:52,220
when the system reaches ambiguity.
1266
00:46:52,220 –> 00:46:54,260
So yes, executives asked for one page,
1267
00:46:54,260 –> 00:46:56,420
what they actually wanted was one place to ask,
1268
00:46:56,420 –> 00:46:57,900
what’s the state of our decisions?
1269
00:46:57,900 –> 00:46:59,700
The co-pilot studio can be that place.
1270
00:46:59,700 –> 00:47:02,580
If you treat it as the interface to a deterministic decision
1271
00:47:02,580 –> 00:47:05,060
engine, not a search for reports.
1272
00:47:05,060 –> 00:47:08,100
Scenario one, set up, revenue forecast variance,
1273
00:47:08,100 –> 00:47:09,540
the classic failure loop.
1274
00:47:09,540 –> 00:47:12,580
Revenue forecast variance is the finance version of an incident.
1275
00:47:12,580 –> 00:47:15,460
It just moves slower with better suits and worse honesty.
1276
00:47:15,460 –> 00:47:17,140
And it repeats because the organization
1277
00:47:17,140 –> 00:47:18,820
keeps treating it as a reporting problem,
1278
00:47:18,820 –> 00:47:20,020
not a decision problem.
1279
00:47:20,020 –> 00:47:21,820
Here’s the classic loop.
1280
00:47:21,820 –> 00:47:24,220
The month closes or the quarter progresses
1281
00:47:24,220 –> 00:47:26,260
and finance produces the variance report.
1282
00:47:26,260 –> 00:47:27,060
Maybe it’s a deck.
1283
00:47:27,060 –> 00:47:29,700
Maybe it’s a power BI page exported to PDF
1284
00:47:29,700 –> 00:47:32,380
because someone still thinks the board is allergic to live data.
1285
00:47:32,380 –> 00:47:34,740
Either way, the packet lands in an executive meeting
1286
00:47:34,740 –> 00:47:36,140
as a statement of state.
1287
00:47:36,140 –> 00:47:39,460
Forecast was X, actual is Y, variance is negative
1288
00:47:39,460 –> 00:47:41,060
and the narrative begins.
1289
00:47:41,060 –> 00:47:42,940
Now the first failure happens immediately.
1290
00:47:42,940 –> 00:47:45,180
The number arrives without obligation.
1291
00:47:45,180 –> 00:47:47,180
Everyone sees the variance, but nobody
1292
00:47:47,180 –> 00:47:49,020
is bound to a pre-committed response.
1293
00:47:49,020 –> 00:47:50,660
So the meeting does what meetings always do
1294
00:47:50,660 –> 00:47:52,260
when the system refuses to decide.
1295
00:47:52,260 –> 00:47:53,260
It debates.
1296
00:47:53,260 –> 00:47:54,980
Sales says it’s pipeline timing.
1297
00:47:54,980 –> 00:47:57,140
Marketing says the leads were lower quality.
1298
00:47:57,140 –> 00:47:59,540
Product sales pricing changes created friction.
1299
00:47:59,540 –> 00:48:01,340
Finance says the expense plan assumed
1300
00:48:01,340 –> 00:48:03,180
a revenue curve that no longer exists.
1301
00:48:03,180 –> 00:48:04,460
Someone says it’s seasonal.
1302
00:48:04,460 –> 00:48:05,980
Someone says it’s one time.
1303
00:48:05,980 –> 00:48:07,340
Someone says the data’s wrong.
1304
00:48:07,340 –> 00:48:09,500
And someone inevitably says let’s monitor it.
1305
00:48:09,500 –> 00:48:10,340
That’s not strategy.
1306
00:48:10,340 –> 00:48:12,100
That’s latency and it’s not accidental.
1307
00:48:12,100 –> 00:48:14,100
The system is designed to produce that outcome
1308
00:48:14,100 –> 00:48:16,020
because the metric is not wired to a trigger
1309
00:48:16,020 –> 00:48:17,620
and owner and an action pathway.
1310
00:48:17,620 –> 00:48:19,980
So humans have to invent the response in real time
1311
00:48:19,980 –> 00:48:22,300
inside a political room under incomplete context.
1312
00:48:22,300 –> 00:48:25,100
Over time that debate loop becomes a behavioral pattern.
1313
00:48:25,100 –> 00:48:27,380
Variance appears, interpretation expands,
1314
00:48:27,380 –> 00:48:30,780
responsibility dilutes, and action slips into the future.
1315
00:48:30,780 –> 00:48:32,300
Then comes the second failure.
1316
00:48:32,300 –> 00:48:35,020
Leading indicators exist, but they don’t count.
1317
00:48:35,020 –> 00:48:36,740
Most organizations have early signals
1318
00:48:36,740 –> 00:48:39,460
that the forecast is drifting, pipeline velocity changes,
1319
00:48:39,460 –> 00:48:42,020
conversion rates, dip, deal aging creeps upward,
1320
00:48:42,020 –> 00:48:44,140
discount rates rise, renewal slippage shows up
1321
00:48:44,140 –> 00:48:45,220
in account notes.
1322
00:48:45,220 –> 00:48:48,740
Sales stages get optimized in CRM to tell a nicer story.
1323
00:48:48,740 –> 00:48:50,740
All of that happens long before the forecast gap
1324
00:48:50,740 –> 00:48:51,740
becomes undeniable.
1325
00:48:51,740 –> 00:48:53,700
But because those signals aren’t owned as rules,
1326
00:48:53,700 –> 00:48:55,540
they’re treated as interesting.
1327
00:48:55,540 –> 00:48:56,820
They become charts people look at,
1328
00:48:56,820 –> 00:48:58,780
not triggers the system enforces.
1329
00:48:58,780 –> 00:49:00,580
So the organization waits for the variance
1330
00:49:00,580 –> 00:49:03,140
to become large enough to be socially undeniable.
1331
00:49:03,140 –> 00:49:04,180
And by the time it is,
1332
00:49:04,180 –> 00:49:06,340
the only actions left are blunt instruments.
1333
00:49:06,340 –> 00:49:09,660
Freeze-spend, cut-head count, cancel programs,
1334
00:49:09,660 –> 00:49:11,220
re-forkast aggressively,
1335
00:49:11,220 –> 00:49:14,460
or invent a new narrative about headwinds and focus.
1336
00:49:14,460 –> 00:49:15,780
That’s the third failure.
1337
00:49:15,780 –> 00:49:18,220
The forecast turns into a story, not a system.
1338
00:49:18,220 –> 00:49:20,620
Forecasting becomes a monthly storytelling ceremony
1339
00:49:20,620 –> 00:49:22,740
where teams negotiate reality under the cover
1340
00:49:22,740 –> 00:49:24,100
of aggregated numbers.
1341
00:49:24,100 –> 00:49:26,460
The model isn’t wrong, the organization is.
1342
00:49:26,460 –> 00:49:29,740
It keeps using forecast variance as a lagging indicator
1343
00:49:29,740 –> 00:49:32,460
then acting surprised when lagging indicators show up late.
1344
00:49:32,460 –> 00:49:34,580
And because the response is late, it’s reactive.
1345
00:49:34,580 –> 00:49:36,420
Reactive actions damage trust.
1346
00:49:36,420 –> 00:49:38,180
They also create collateral damage.
1347
00:49:38,180 –> 00:49:40,500
Marketing gets whiplash, sales gets pressure
1348
00:49:40,500 –> 00:49:42,380
that turns into deal quality decay
1349
00:49:42,380 –> 00:49:43,980
and finance loses credibility
1350
00:49:43,980 –> 00:49:45,980
because it looks like it can’t see the future,
1351
00:49:45,980 –> 00:49:47,900
which is unfair because it usually can.
1352
00:49:47,900 –> 00:49:49,260
It just can’t enforce anything.
1353
00:49:49,260 –> 00:49:51,540
Now zoom out to the meta problem.
1354
00:49:51,540 –> 00:49:54,660
Forecast variance creates strategy/thrash.
1355
00:49:54,660 –> 00:49:56,420
Executives can’t commit to initiatives
1356
00:49:56,420 –> 00:49:58,580
because the financial narrative changes every month.
1357
00:49:58,580 –> 00:50:00,700
Teams can’t plan because priorities shift
1358
00:50:00,700 –> 00:50:02,780
with the latest variance explanation
1359
00:50:02,780 –> 00:50:05,340
and each cycle burns organizational energy.
1360
00:50:05,340 –> 00:50:08,140
More reconciliation, more follow-up analysis,
1361
00:50:08,140 –> 00:50:10,340
more quick-deep dives, more slide updates,
1362
00:50:10,340 –> 00:50:12,260
more meetings whose only output is a decision
1363
00:50:12,260 –> 00:50:13,700
to have more meetings.
1364
00:50:13,700 –> 00:50:16,740
This is why finance leaders keep asking for a single page.
1365
00:50:16,740 –> 00:50:19,460
They think compression or visibility will compress the cycle.
1366
00:50:19,460 –> 00:50:20,260
It won’t.
1367
00:50:20,260 –> 00:50:23,220
A one-page KPI overview might show the variance sooner
1368
00:50:23,220 –> 00:50:24,620
with nicer formatting.
1369
00:50:24,620 –> 00:50:26,260
But it doesn’t change the response pattern.
1370
00:50:26,260 –> 00:50:27,580
It doesn’t assign ownership.
1371
00:50:27,580 –> 00:50:28,980
It doesn’t pre-commit actions.
1372
00:50:28,980 –> 00:50:30,380
It doesn’t enforce time windows.
1373
00:50:30,380 –> 00:50:31,460
It doesn’t store state.
1374
00:50:31,460 –> 00:50:33,380
It doesn’t measure whether interventions worked.
1375
00:50:33,380 –> 00:50:34,820
So the loop persists.
1376
00:50:34,820 –> 00:50:36,020
Variance report.
1377
00:50:36,020 –> 00:50:37,020
Meeting debate.
1378
00:50:37,020 –> 00:50:38,220
Delayed adjustment.
1379
00:50:38,220 –> 00:50:39,580
Next month repeat.
1380
00:50:39,580 –> 00:50:41,780
And the most expensive part isn’t the bad month.
1381
00:50:41,780 –> 00:50:44,300
It’s the two weeks of organizational indecision
1382
00:50:44,300 –> 00:50:45,780
where everyone can see the drift
1383
00:50:45,780 –> 00:50:47,940
but nobody is structurally obligated to act.
1384
00:50:47,940 –> 00:50:51,060
That’s the real KPI decision latency.
1385
00:50:51,060 –> 00:50:53,260
Now the only way out is to treat forecast variance
1386
00:50:53,260 –> 00:50:56,380
like an engineered decision surface, not a finance artifact,
1387
00:50:56,380 –> 00:50:58,980
not a slide.
1388
00:50:58,980 –> 00:51:01,780
A decision surface where leading indicators trigger
1389
00:51:01,780 –> 00:51:05,580
deterministic obligations, where actions execute by design,
1390
00:51:05,580 –> 00:51:07,340
and where exceptions are visible instead
1391
00:51:07,340 –> 00:51:08,900
of smuggled in as narratives.
1392
00:51:08,900 –> 00:51:10,380
So next, that’s what we’re going to do.
1393
00:51:10,380 –> 00:51:13,020
We’ll convert forecast variance from a monthly argument
1394
00:51:13,020 –> 00:51:16,460
into a deterministic decision design, triggers, owners,
1395
00:51:16,460 –> 00:51:18,700
actions, time windows, and feedback.
1396
00:51:18,700 –> 00:51:20,780
And it will feel uncomfortable because it removes
1397
00:51:20,780 –> 00:51:23,780
the organization’s favorite capability, delaying reality
1398
00:51:23,780 –> 00:51:25,620
with meetings.
1399
00:51:25,620 –> 00:51:27,420
Scenario one decision design.
1400
00:51:27,420 –> 00:51:30,340
Forecast variance as triggers owners actions.
1401
00:51:30,340 –> 00:51:31,620
So let’s redesign it.
1402
00:51:31,620 –> 00:51:34,660
Not as a prettier variance page as a deterministic decision
1403
00:51:34,660 –> 00:51:36,300
surface that treats forecast drift
1404
00:51:36,300 –> 00:51:37,980
like a controllable failure mode.
1405
00:51:37,980 –> 00:51:39,740
Start with the trigger because without the trigger,
1406
00:51:39,740 –> 00:51:40,820
you’re back to vibes.
1407
00:51:40,820 –> 00:51:43,340
The trigger can’t be variance is getting worse.
1408
00:51:43,340 –> 00:51:45,620
That’s a future argument disguised as monitoring.
1409
00:51:45,620 –> 00:51:47,780
It has to be a condition the system can evaluate
1410
00:51:47,780 –> 00:51:50,780
the same way every time, threshold, duration, and context.
1411
00:51:50,780 –> 00:51:53,580
For example, forecast variance below negative 7%
1412
00:51:53,580 –> 00:51:56,460
for 10 business days evaluated on a rolling basis
1413
00:51:56,460 –> 00:51:58,660
scope to a segment that actually matters.
1414
00:51:58,660 –> 00:52:01,620
Not the whole business where one product masks another.
1415
00:52:01,620 –> 00:52:04,980
And not a region so broad, it becomes politically safe.
1416
00:52:04,980 –> 00:52:07,620
Pick the scope that corresponds to the owner’s authority.
1417
00:52:07,620 –> 00:52:10,300
If the VP of rev-ops can actually influence enterprise
1418
00:52:10,300 –> 00:52:13,140
pipeline in EMIA, then the trigger should fire there.
1419
00:52:13,140 –> 00:52:15,260
If they can’t, you’re designing a trigger
1420
00:52:15,260 –> 00:52:17,660
that produces performative accountability.
1421
00:52:17,660 –> 00:52:19,300
Now here’s the part people avoid.
1422
00:52:19,300 –> 00:52:21,220
The trigger needs a pre-breach signal.
1423
00:52:21,220 –> 00:52:24,020
Waiting for monthly variance is like waiting for a service
1424
00:52:24,020 –> 00:52:25,940
outage report to start incident response.
1425
00:52:25,940 –> 00:52:28,300
You want leading indicators that predict drift
1426
00:52:28,300 –> 00:52:30,460
early enough for intervention to be cheap.
1427
00:52:30,460 –> 00:52:31,940
So define a second trigger set.
1428
00:52:31,940 –> 00:52:34,940
Pipeline velocity drop beyond the defined band, conversion rate
1429
00:52:34,940 –> 00:52:37,780
decay, deal aging, crossing a threshold in late stage,
1430
00:52:37,780 –> 00:52:40,100
renewal, slippage count, exceeding a cap.
1431
00:52:40,100 –> 00:52:42,980
Each of those can be evaluated with the same determinism,
1432
00:52:42,980 –> 00:52:45,420
threshold, duration, and scope.
1433
00:52:45,420 –> 00:52:47,820
And yes, you’ll get pushback because leading indicators
1434
00:52:47,820 –> 00:52:50,020
remove the ability to claim surprise.
1435
00:52:50,020 –> 00:52:51,180
Next is ownership lock.
1436
00:52:51,180 –> 00:52:53,500
And this is where the architecture turns from analytics
1437
00:52:53,500 –> 00:52:55,340
into organizational design.
1438
00:52:55,340 –> 00:52:58,220
One role owns the variance trigger, one not finance,
1439
00:52:58,220 –> 00:53:01,100
not sales leadership as a concept, not the business.
1440
00:53:01,100 –> 00:53:03,540
A specific accountable function like VP of revenue
1441
00:53:03,540 –> 00:53:04,980
operations because rev-ops sits
1442
00:53:04,980 –> 00:53:07,780
at the intersection of pipeline mechanics, forecasting
1443
00:53:07,780 –> 00:53:09,740
process and operational levers.
1444
00:53:09,740 –> 00:53:12,300
Finance becomes steward of definition and audit posture,
1445
00:53:12,300 –> 00:53:13,900
not the execution owner.
1446
00:53:13,900 –> 00:53:15,780
Sales becomes a participant in actions,
1447
00:53:15,780 –> 00:53:18,140
not the place where ownership goes to dissolve.
1448
00:53:18,140 –> 00:53:20,060
And ownership has to be more than a name.
1449
00:53:20,060 –> 00:53:21,140
It’s an obligation.
1450
00:53:21,140 –> 00:53:22,660
When the trigger fires, that role
1451
00:53:22,660 –> 00:53:25,580
must execute the response pathway or file an exception
1452
00:53:25,580 –> 00:53:27,860
that’s visible, timestamped, and reviewable.
1453
00:53:27,860 –> 00:53:29,140
Exceptions are allowed.
1454
00:53:29,140 –> 00:53:31,060
Invisible exceptions are not.
1455
00:53:31,060 –> 00:53:32,540
Now pre-committed actions.
1456
00:53:32,540 –> 00:53:34,540
This is the real divider between reporting
1457
00:53:34,540 –> 00:53:36,100
culture and decision culture.
1458
00:53:36,100 –> 00:53:37,940
You don’t want a meeting to decide what to do.
1459
00:53:37,940 –> 00:53:40,580
You want a meeting only when the system hits an exception state.
1460
00:53:40,580 –> 00:53:42,580
So you define action pathways in advance
1461
00:53:42,580 –> 00:53:44,180
aligned to the type of trigger.
1462
00:53:44,180 –> 00:53:47,180
If the trigger is forecast variance exceeding the threshold,
1463
00:53:47,180 –> 00:53:49,100
the pathway could be one.
1464
00:53:49,100 –> 00:53:50,700
Apply, spend guardrails.
1465
00:53:50,700 –> 00:53:53,060
Freeze discretionary spend above a defined threshold
1466
00:53:53,060 –> 00:53:55,140
until variance returns inside tolerance
1467
00:53:55,140 –> 00:53:57,980
with an explicit exception mechanism for strategic items.
1468
00:53:57,980 –> 00:53:59,260
Not we should be careful.
1469
00:53:59,260 –> 00:54:01,260
A guardrail, the system enforces.
1470
00:54:01,260 –> 00:54:04,180
Two, launch a pipeline acceleration sprint.
1471
00:54:04,180 –> 00:54:07,180
A time-boxed playbook, review-late-stage deal aging,
1472
00:54:07,180 –> 00:54:09,740
and force next step dates, remove stall deals
1473
00:54:09,740 –> 00:54:12,380
from late-stage and trigger executive sponsor outreach
1474
00:54:12,380 –> 00:54:13,780
for a defined subset.
1475
00:54:13,780 –> 00:54:16,740
Again, not a suggestion, a playbook.
1476
00:54:16,740 –> 00:54:19,300
Three, reallocation rules.
1477
00:54:19,300 –> 00:54:21,100
Shift a defined portion of paid budget
1478
00:54:21,100 –> 00:54:23,660
toward channels with demonstrated conversion efficiency
1479
00:54:23,660 –> 00:54:25,780
or pause under performing campaigns.
1480
00:54:25,780 –> 00:54:27,220
The amount is defined in advance,
1481
00:54:27,220 –> 00:54:30,460
so people can’t negotiate the response under pressure.
1482
00:54:30,460 –> 00:54:32,740
Four, forecast reset conditions.
1483
00:54:32,740 –> 00:54:34,500
If the system detects persistent bias,
1484
00:54:34,500 –> 00:54:37,220
it forces a re-forecast using agreed methodology.
1485
00:54:37,220 –> 00:54:38,820
Not because the number is embarrassing,
1486
00:54:38,820 –> 00:54:40,180
because the model is drifting,
1487
00:54:40,180 –> 00:54:43,220
and each action has to be executable, not poetic.
1488
00:54:43,220 –> 00:54:45,100
If the action is improved pipeline quality,
1489
00:54:45,100 –> 00:54:48,500
that’s not an action that’s a wish, then time constraints.
1490
00:54:48,500 –> 00:54:51,460
The window has to match the risk curve, not the calendar.
1491
00:54:51,460 –> 00:54:52,980
For leading indicator triggers,
1492
00:54:52,980 –> 00:54:55,180
the window might be 48 hours.
1493
00:54:55,180 –> 00:54:58,180
Acknowledge, assign, execute playbook start.
1494
00:54:58,180 –> 00:54:59,380
For variance breach triggers,
1495
00:54:59,380 –> 00:55:01,940
it might be 24 hours for guardrails and 72
1496
00:55:01,940 –> 00:55:03,500
for the first stabilization check.
1497
00:55:03,500 –> 00:55:06,620
Time windows change behavior because they make delay measurable.
1498
00:55:06,620 –> 00:55:08,300
If the owner misses the window,
1499
00:55:08,300 –> 00:55:11,620
that’s a breach of the decision system, not a busy week.
1500
00:55:11,620 –> 00:55:12,740
Now the feedback loop.
1501
00:55:12,740 –> 00:55:15,060
You don’t measure success by whether people felt busy.
1502
00:55:15,060 –> 00:55:17,300
You measure intervention efficacy.
1503
00:55:17,300 –> 00:55:19,580
Did forecast variance move back inside tolerance
1504
00:55:19,580 –> 00:55:20,620
within the committed time?
1505
00:55:20,620 –> 00:55:22,820
Did forecast error reduce over the next cycle?
1506
00:55:22,820 –> 00:55:24,300
Did pipeline velocity recover?
1507
00:55:24,300 –> 00:55:25,620
Did conversion stabilize?
1508
00:55:25,620 –> 00:55:26,700
And what was the cost?
1509
00:55:26,700 –> 00:55:28,620
Did discounting spike, did churn increase?
1510
00:55:28,620 –> 00:55:30,260
Did sales cycle time extend?
1511
00:55:30,260 –> 00:55:33,020
This is where you prevent gaming by pairing metrics.
1512
00:55:33,020 –> 00:55:36,700
Recovery paired with margin, acceleration paired with quality,
1513
00:55:36,700 –> 00:55:37,940
speed paired with risk.
1514
00:55:37,940 –> 00:55:40,620
And all of this gets recorded as state, trigger fired,
1515
00:55:40,620 –> 00:55:43,100
owner assigned, action path executed,
1516
00:55:43,100 –> 00:55:45,580
exception approved or denied, outcome measured,
1517
00:55:45,580 –> 00:55:46,900
rule version recorded.
1518
00:55:46,900 –> 00:55:49,100
Now you have something finance leaders can defend
1519
00:55:49,100 –> 00:55:51,060
and something executives can actually run
1520
00:55:51,060 –> 00:55:53,100
because the forecast stops being a story.
1521
00:55:53,100 –> 00:55:54,580
It becomes a managed system.
1522
00:55:54,580 –> 00:55:55,540
And the payoff is simple.
1523
00:55:55,540 –> 00:55:57,620
The organization doesn’t wait two weeks to decide
1524
00:55:57,620 –> 00:55:58,740
whether reality is real.
1525
00:55:58,740 –> 00:56:01,660
It responds inside 48 hours by design.
1526
00:56:01,660 –> 00:56:04,580
Scenario two set up, IT incident, SLA.
1527
00:56:04,580 –> 00:56:07,260
The dashboard that reports failure after it happens.
1528
00:56:07,260 –> 00:56:08,540
Now take the same failure pattern
1529
00:56:08,540 –> 00:56:10,140
and move it from finance into IT.
1530
00:56:10,140 –> 00:56:12,180
It gets uglier, faster and more measurable
1531
00:56:12,180 –> 00:56:14,660
because incident SLA is our way organizations pretend
1532
00:56:14,660 –> 00:56:17,060
they have control right up until the clock runs out
1533
00:56:17,060 –> 00:56:19,300
and the dashboard politely confirms the damage.
1534
00:56:19,300 –> 00:56:21,700
Here’s the classic set up.
1535
00:56:21,700 –> 00:56:25,220
A service desk or operations team produces an SLA dashboard,
1536
00:56:25,220 –> 00:56:27,620
usually weekly, sometimes daily.
1537
00:56:27,620 –> 00:56:30,660
It shows compliance percentages, breach counts,
1538
00:56:30,660 –> 00:56:32,940
maybe a trend line if someone feels ambitious.
1539
00:56:32,940 –> 00:56:34,660
Leadership sees green, breathes out.
1540
00:56:34,660 –> 00:56:37,180
Leadership sees red, schedules or conversation.
1541
00:56:37,180 –> 00:56:38,300
And that’s the problem.
1542
00:56:38,300 –> 00:56:41,100
An SLA dashboard is typically a post-mortem artifact
1543
00:56:41,100 –> 00:56:42,620
disguised as a control system.
1544
00:56:42,620 –> 00:56:45,420
It tells you what happened after the only moment that mattered.
1545
00:56:45,420 –> 00:56:47,460
The moment you still could have prevented the breach.
1546
00:56:47,460 –> 00:56:48,900
So it becomes theater.
1547
00:56:48,900 –> 00:56:50,780
Not because the ops team is lazy
1548
00:56:50,780 –> 00:56:53,420
because the architecture is designed to observe failure,
1549
00:56:53,420 –> 00:56:54,420
not prevented.
1550
00:56:54,420 –> 00:56:56,060
The first failure loop is timing.
1551
00:56:56,060 –> 00:56:58,580
If your primary SLA visibility cadence is weekly,
1552
00:56:58,580 –> 00:57:00,460
you’ve already accepted that the organization
1553
00:57:00,460 –> 00:57:03,900
will discover systemic breach patterns on a calendar schedule.
1554
00:57:03,900 –> 00:57:06,780
But outages and degradations don’t care about your calendar.
1555
00:57:06,780 –> 00:57:09,820
They follow physics, load, change and human error.
1556
00:57:09,820 –> 00:57:12,060
So the system produces an incident, the clock starts,
1557
00:57:12,060 –> 00:57:14,980
the team does what it can, and then the dashboard shows up
1558
00:57:14,980 –> 00:57:16,220
later to narrate the outcome.
1559
00:57:16,220 –> 00:57:18,700
Everyone nods, everyone agrees it can’t happen again,
1560
00:57:18,700 –> 00:57:20,660
then it happens again, because nothing changed
1561
00:57:20,660 –> 00:57:22,140
in the decision mechanics.
1562
00:57:22,140 –> 00:57:24,540
Second failure loop, severity debates.
1563
00:57:24,540 –> 00:57:27,580
Most teams act like severity is an objective classification.
1564
00:57:27,580 –> 00:57:28,500
It isn’t.
1565
00:57:28,500 –> 00:57:31,060
In many organizations, severity becomes a negotiation
1566
00:57:31,060 –> 00:57:32,740
because severity controls optics.
1567
00:57:32,740 –> 00:57:34,060
It controls who gets paged.
1568
00:57:34,060 –> 00:57:35,220
It controls escalation.
1569
00:57:35,220 –> 00:57:37,140
It controls whether leadership gets involved.
1570
00:57:37,140 –> 00:57:39,900
It controls whether a breach becomes a learning opportunity
1571
00:57:39,900 –> 00:57:41,380
or a performance issue.
1572
00:57:41,380 –> 00:57:42,300
So what happens?
1573
00:57:42,300 –> 00:57:44,540
An incident occurs, someone labels it one way,
1574
00:57:44,540 –> 00:57:47,220
someone else challenges it, someone waits for more information.
1575
00:57:47,220 –> 00:57:50,100
And the SLA clock keeps running while the organization
1576
00:57:50,100 –> 00:57:51,620
argues about the label.
1577
00:57:51,620 –> 00:57:53,420
That debate is an entropy generator.
1578
00:57:53,420 –> 00:57:56,580
And dashboards love it because dashboards only need a final label.
1579
00:57:56,580 –> 00:57:58,860
They don’t care how much time you burned getting there.
1580
00:57:58,860 –> 00:58:01,340
Third failure loop, manual escalation.
1581
00:58:01,340 –> 00:58:03,860
In reporting culture, escalation is a human behavior,
1582
00:58:03,860 –> 00:58:05,180
not a system behavior.
1583
00:58:05,180 –> 00:58:07,220
Someone has to notice the risk, decided serious,
1584
00:58:07,220 –> 00:58:09,020
find the right person, get their attention,
1585
00:58:09,020 –> 00:58:11,100
and then convince them to care at the right level.
1586
00:58:11,100 –> 00:58:14,020
So response times become random, not because people don’t care,
1587
00:58:14,020 –> 00:58:16,300
because the system has no deterministic rooting
1588
00:58:16,300 –> 00:58:18,660
and no pre-committed escalation tiers.
1589
00:58:18,660 –> 00:58:20,700
It relies on social networks and heroics,
1590
00:58:20,700 –> 00:58:22,700
and heroics don’t scale.
1591
00:58:22,700 –> 00:58:25,380
They create a false sense of competence
1592
00:58:25,380 –> 00:58:28,620
that collapses the minute the right person is on vacation.
1593
00:58:28,620 –> 00:58:32,380
Fourth failure loop, root cause after breach.
1594
00:58:32,380 –> 00:58:34,620
A lot of organizations do root cause analysis
1595
00:58:34,620 –> 00:58:37,140
the way some people do taxes, late, painful,
1596
00:58:37,140 –> 00:58:39,140
and primarily for compliance optics.
1597
00:58:39,140 –> 00:58:40,860
They investigate after the SLA breach
1598
00:58:40,860 –> 00:58:42,620
because the breach forced visibility.
1599
00:58:42,620 –> 00:58:45,980
But the operational truth is that the breach was predictable earlier.
1600
00:58:45,980 –> 00:58:48,660
Q build up, missing ownership, stalled handoffs,
1601
00:58:48,660 –> 00:58:51,380
unclear severity, insufficient on-call coverage,
1602
00:58:51,380 –> 00:58:54,100
lack of runbooks, whatever the real cause is.
1603
00:58:54,100 –> 00:58:56,460
The dashboard wasn’t blind, it was just too late.
1604
00:58:56,460 –> 00:58:59,380
And this is where the KPI request shows up in IT language.
1605
00:58:59,380 –> 00:59:03,140
Leadership will say, “I want a one-page view of incidents and SLA.”
1606
00:59:03,140 –> 00:59:04,260
And what they mean is,
1607
00:59:04,260 –> 00:59:06,220
“I want to know that breach prevention is guaranteed
1608
00:59:06,220 –> 00:59:07,700
without me having to chase people.
1609
00:59:07,700 –> 00:59:08,820
They want a control plane.
1610
00:59:08,820 –> 00:59:10,420
They want determinism.
1611
00:59:10,420 –> 00:59:12,540
They want the same thing finance wanted
1612
00:59:12,540 –> 00:59:15,100
when a condition is met, something is already in motion.”
1613
00:59:15,100 –> 00:59:18,100
So what does the one-page become in a decision engine world?
1614
00:59:18,100 –> 00:59:20,100
It becomes a breach prevention surface,
1615
00:59:20,100 –> 00:59:21,740
not a breach documentation surface.
1616
00:59:21,740 –> 00:59:23,820
The KPI isn’t the SLA percentage.
1617
00:59:23,820 –> 00:59:24,660
That’s a metric.
1618
00:59:24,660 –> 00:59:27,380
The KPI is the rule when breach risk crosses a threshold,
1619
00:59:27,380 –> 00:59:29,580
escalation happens, playbooks launch,
1620
00:59:29,580 –> 00:59:31,660
and ownership locks automatically,
1621
00:59:31,660 –> 00:59:34,540
which means the setup problem in IT isn’t lack of dashboards.
1622
00:59:34,540 –> 00:59:36,580
It’s the fact that the organization has outsourced
1623
00:59:36,580 –> 00:59:39,820
time-critical rooting to meetings, inboxes, and opinions.
1624
00:59:39,820 –> 00:59:42,940
And unlike finance, IT doesn’t give you two weeks to argue.
1625
00:59:42,940 –> 00:59:45,060
The clock is visible, customers feel it.
1626
00:59:45,060 –> 00:59:46,700
And once the SLA is breached,
1627
00:59:46,700 –> 00:59:48,780
you don’t get to retroactively respond faster.
1628
00:59:48,780 –> 00:59:52,140
So the SLA dashboard becomes the ultimate wallpaper chart.
1629
00:59:52,140 –> 00:59:54,780
Accurate, well-formatted, and functionally useless
1630
00:59:54,780 –> 00:59:55,940
at the moment that matters.
1631
00:59:55,940 –> 00:59:57,260
Now the redesign has to start
1632
00:59:57,260 –> 00:59:59,500
where reporting culture refuses to go.
1633
00:59:59,500 –> 01:00:02,820
Deterministic severity, deterministic breach countdowns,
1634
01:00:02,820 –> 01:00:04,460
deterministic escalation tiers,
1635
01:00:04,460 –> 01:00:06,380
and pre-committed remediation playbooks
1636
01:00:06,380 –> 01:00:08,300
that launch before breach, not after.
1637
01:00:08,300 –> 01:00:09,740
That’s scenario two.
1638
01:00:09,740 –> 01:00:12,620
And it’s where the architecture either grows up into enforcement
1639
01:00:12,620 –> 01:00:14,620
or stays trapped in visibility
1640
01:00:14,620 –> 01:00:16,940
and keeps calling it control.
1641
01:00:16,940 –> 01:00:19,540
SLA two decision design, SLA compliance
1642
01:00:19,540 –> 01:00:21,660
as a deterministic escalation engine.
1643
01:00:21,660 –> 01:00:23,660
So let’s redesign SLA compliance
1644
01:00:23,660 –> 01:00:25,660
the same way we redesigned forecast variants
1645
01:00:25,660 –> 01:00:27,500
as a deterministic escalation engine,
1646
01:00:27,500 –> 01:00:29,540
not a weekly dashboard artifact.
1647
01:00:29,540 –> 01:00:32,340
Start with the piece, everyone pretends is objective,
1648
01:00:32,340 –> 01:00:33,980
but usually isn’t.
1649
01:00:33,980 –> 01:00:35,580
Severity.
1650
01:00:35,580 –> 01:00:37,660
If severity classification requires a meeting,
1651
01:00:37,660 –> 01:00:38,580
you’ve already lost.
1652
01:00:38,580 –> 01:00:41,780
The SLA clock doesn’t pause while you negotiate optics.
1653
01:00:41,780 –> 01:00:44,220
So severity has to be deterministic enough
1654
01:00:44,220 –> 01:00:46,420
that the first 15 minutes of an incident
1655
01:00:46,420 –> 01:00:48,500
aren’t spent arguing about what to call it.
1656
01:00:48,500 –> 01:00:49,700
That doesn’t mean it’s perfect.
1657
01:00:49,700 –> 01:00:50,900
That means it’s rule-based.
1658
01:00:50,900 –> 01:00:53,340
Severity becomes a function of observable attributes
1659
01:00:53,340 –> 01:00:56,900
you can commit to, customer impact scope, service criticality
1660
01:00:56,900 –> 01:01:00,900
tier, data loss risk, security exposure, and time to breach.
1661
01:01:00,900 –> 01:01:03,060
You don’t let how loud the stakeholder is
1662
01:01:03,060 –> 01:01:05,980
become an input variable, even though it will try.
1663
01:01:05,980 –> 01:01:07,260
And here’s the part most people miss.
1664
01:01:07,260 –> 01:01:08,420
Severity isn’t a label.
1665
01:01:08,420 –> 01:01:09,500
It’s a rooting decision.
1666
01:01:09,500 –> 01:01:12,500
Severity determines who gets paged, what playbook starts,
1667
01:01:12,500 –> 01:01:14,140
what communications are pre-approved,
1668
01:01:14,140 –> 01:01:15,900
and how fast escalation happens.
1669
01:01:15,900 –> 01:01:18,620
So you define severity like you define firewall rules,
1670
01:01:18,620 –> 01:01:20,620
explicit conditions, explicit consequences
1671
01:01:20,620 –> 01:01:21,740
and no hidden exceptions.
1672
01:01:21,740 –> 01:01:23,460
Now the trigger.
1673
01:01:23,460 –> 01:01:26,780
The KPI in this design is not SLA compliance this week.
1674
01:01:26,780 –> 01:01:27,860
That’s a tombstone.
1675
01:01:27,860 –> 01:01:29,700
The KPI is breach risk.
1676
01:01:29,700 –> 01:01:32,020
So the trigger fires before the breach, not after.
1677
01:01:32,020 –> 01:01:34,780
For example, it is, if time remaining to SLA
1678
01:01:34,780 –> 01:01:37,020
is under 30 minutes for a CV-1 incident
1679
01:01:37,020 –> 01:01:39,380
and status is not mitigation in progress,
1680
01:01:39,380 –> 01:01:42,700
escalate to tier two and open the major incident bridge.
1681
01:01:42,700 –> 01:01:45,860
Or if time remaining to SLA is under two hours for 7-2
1682
01:01:45,860 –> 01:01:47,980
and no owner has acknowledged within 15 minutes
1683
01:01:47,980 –> 01:01:49,500
escalate to the service owner,
1684
01:01:49,500 –> 01:01:52,020
the trigger includes duration because flapping happens.
1685
01:01:52,020 –> 01:01:53,940
One noisy update shouldn’t wake people up,
1686
01:01:53,940 –> 01:01:55,500
but a sustained risk condition should.
1687
01:01:55,500 –> 01:01:56,780
And you lock context constraints
1688
01:01:56,780 –> 01:01:58,620
so you don’t drown in false escalations,
1689
01:01:58,620 –> 01:02:00,740
only certain services, only certain customers,
1690
01:02:00,740 –> 01:02:03,420
only incidents that meet the rules eligibility criteria.
1691
01:02:03,420 –> 01:02:06,500
Otherwise you just teach people to ignore the system
1692
01:02:06,500 –> 01:02:10,140
which is how entropy reenters through culture, then ownership.
1693
01:02:10,140 –> 01:02:12,100
This is where IT loves committees.
1694
01:02:12,100 –> 01:02:13,260
The team owns it.
1695
01:02:13,260 –> 01:02:14,260
No, the team doesn’t.
1696
01:02:14,260 –> 01:02:17,140
A deterministic engine assigns one accountable role
1697
01:02:17,140 –> 01:02:20,340
per incident state, incident commander, service owner,
1698
01:02:20,340 –> 01:02:22,340
on-call engineer, communications owner,
1699
01:02:22,340 –> 01:02:23,660
not because they do every task,
1700
01:02:23,660 –> 01:02:25,900
but because the system needs one throat to choke
1701
01:02:25,900 –> 01:02:27,140
when the clock is burning.
1702
01:02:27,140 –> 01:02:28,820
Ownership also can’t be static.
1703
01:02:28,820 –> 01:02:31,620
It has to resolve a runtime who is on call right now,
1704
01:02:31,620 –> 01:02:32,940
who is primary and secondary,
1705
01:02:32,940 –> 01:02:34,740
what escalation tier is active.
1706
01:02:34,740 –> 01:02:36,460
If your ownership model relies on someone
1707
01:02:36,460 –> 01:02:38,180
updating a spreadsheet, it will drift.
1708
01:02:38,180 –> 01:02:39,060
It always does.
1709
01:02:39,060 –> 01:02:41,940
Now the pre-committed actions.
1710
01:02:41,940 –> 01:02:46,180
This is where you stop pretending that escalate is an action.
1711
01:02:46,180 –> 01:02:47,740
Escalation is a rooting mechanism.
1712
01:02:47,740 –> 01:02:49,460
The action is what happens once rooted.
1713
01:02:49,460 –> 01:02:51,820
So you define playbooks that are not documents
1714
01:02:51,820 –> 01:02:53,700
in a SharePoint folder, nobody opens,
1715
01:02:53,700 –> 01:02:56,260
but operational sequences the system can launch.
1716
01:02:56,260 –> 01:02:58,900
For SaveOne, open a bridge automatically,
1717
01:02:58,900 –> 01:03:00,700
page tier one and tier two,
1718
01:03:00,700 –> 01:03:03,180
create the incident record in the decision ledger,
1719
01:03:03,180 –> 01:03:04,860
assign incident commander,
1720
01:03:04,860 –> 01:03:06,500
start the breach countdown timer,
1721
01:03:06,500 –> 01:03:09,460
and post a pre-approved customer impact message template
1722
01:03:09,460 –> 01:03:10,500
for review.
1723
01:03:10,500 –> 01:03:11,900
For SaveTwo, page the on-call,
1724
01:03:11,900 –> 01:03:13,660
create tasks for the first diagnostics,
1725
01:03:13,660 –> 01:03:15,860
require acknowledgement and schedule escalation
1726
01:03:15,860 –> 01:03:18,780
if acknowledgement doesn’t happen inside the defined window.
1727
01:03:18,780 –> 01:03:19,820
And you build guardrails.
1728
01:03:19,820 –> 01:03:21,620
If the incident touches regulated data,
1729
01:03:21,620 –> 01:03:23,980
the communications pathway requires human approval.
1730
01:03:23,980 –> 01:03:25,940
If it doesn’t, the system can proceed
1731
01:03:25,940 –> 01:03:27,820
with pre-approved internal notifications
1732
01:03:27,820 –> 01:03:29,140
and customer updates.
1733
01:03:29,140 –> 01:03:32,060
Again, approvals only wear risk demands human judgment.
1734
01:03:32,060 –> 01:03:34,020
Then time constraints, which is the whole point
1735
01:03:34,020 –> 01:03:35,500
of SLA engineering.
1736
01:03:35,500 –> 01:03:37,140
You don’t wait for end of day.
1737
01:03:37,140 –> 01:03:38,740
You don’t wait for the weekly review.
1738
01:03:38,740 –> 01:03:42,340
You codify escalation tiers as time-based system law.
1739
01:03:42,340 –> 01:03:44,220
At T-minus 60 minutes,
1740
01:03:44,220 –> 01:03:46,380
escalate from tier one to tier two.
1741
01:03:46,380 –> 01:03:50,100
At T-minus 30, incident commander becomes mandatory.
1742
01:03:50,100 –> 01:03:52,620
At T-minus 15, leadership notification triggers
1743
01:03:52,620 –> 01:03:54,900
and customer comes prep becomes non-optional.
1744
01:03:54,900 –> 01:03:56,580
And if the system hits T-minus zero,
1745
01:03:56,580 –> 01:03:59,460
that’s not a KPI turning red, that’s an architectural failure.
1746
01:03:59,460 –> 01:04:01,180
Either the trigger thresholds were wrong,
1747
01:04:01,180 –> 01:04:03,700
the playbook was missing or ownership didn’t execute,
1748
01:04:03,700 –> 01:04:05,180
and the ledger must record which one,
1749
01:04:05,180 –> 01:04:07,580
because otherwise you’ll blame process and fix nothing.
1750
01:04:07,580 –> 01:04:08,740
Finally, the feedback loop,
1751
01:04:08,740 –> 01:04:11,860
because IT loves post mortems and still manages not to learn.
1752
01:04:11,860 –> 01:04:13,580
So you measure breach prevention rate,
1753
01:04:13,580 –> 01:04:14,580
not just breach count.
1754
01:04:14,580 –> 01:04:15,660
You measure M-T-T-R,
1755
01:04:15,660 –> 01:04:17,820
but also time to acknowledge, time to escalate
1756
01:04:17,820 –> 01:04:19,660
and time spent in severity disputes.
1757
01:04:19,660 –> 01:04:21,460
Those are the hidden entropy metrics.
1758
01:04:21,460 –> 01:04:23,140
You measure false escalation rate,
1759
01:04:23,140 –> 01:04:25,300
because too many false positives trains people
1760
01:04:25,300 –> 01:04:26,740
to ignore the system.
1761
01:04:26,740 –> 01:04:29,060
And you tune the trigger rules and context constraints
1762
01:04:29,060 –> 01:04:31,180
based on observed outcomes, not on opinions.
1763
01:04:31,180 –> 01:04:32,340
This is the shift.
1764
01:04:32,340 –> 01:04:35,020
SLA compliance stops being a number you report
1765
01:04:35,020 –> 01:04:37,100
and becomes a capability you operate.
1766
01:04:37,100 –> 01:04:39,620
Same condition, same classification, same routing,
1767
01:04:39,620 –> 01:04:42,060
same escalation tiers, same playbook launch,
1768
01:04:42,060 –> 01:04:44,380
same ledger evidence, and when leadership asks,
1769
01:04:44,380 –> 01:04:45,460
are we in trouble?
1770
01:04:45,460 –> 01:04:46,780
They don’t get a chart.
1771
01:04:46,780 –> 01:04:48,020
They get posture.
1772
01:04:48,020 –> 01:04:50,620
Which incidents are inside breach risk thresholds
1773
01:04:50,620 –> 01:04:52,940
which are already escalated, which actions are pending
1774
01:04:52,940 –> 01:04:53,900
and who is accountable?
1775
01:04:53,900 –> 01:04:55,700
That’s what one page was supposed to mean.
1776
01:04:55,700 –> 01:04:57,860
Deterministic versus probabilistic,
1777
01:04:57,860 –> 01:05:00,380
where AI belongs and where it is banned.
1778
01:05:00,380 –> 01:05:02,620
Now we hit the part everyone wants to skip too,
1779
01:05:02,620 –> 01:05:05,260
because it sounds modern and feels like progress, AI.
1780
01:05:05,260 –> 01:05:06,940
And this is where organizations create
1781
01:05:06,940 –> 01:05:09,020
the most expensive form of entropy.
1782
01:05:09,020 –> 01:05:10,940
They take a system that already can’t decide
1783
01:05:10,940 –> 01:05:12,940
and they add a probabilistic narrator to it.
1784
01:05:12,940 –> 01:05:13,940
That’s not intelligence.
1785
01:05:13,940 –> 01:05:15,900
That’s conditional chaos with better grammar.
1786
01:05:15,900 –> 01:05:18,180
So here’s the rule that makes this usable.
1787
01:05:18,180 –> 01:05:20,020
You build a deterministic core
1788
01:05:20,020 –> 01:05:22,940
and you allow probabilistic AI only at the edges.
1789
01:05:22,940 –> 01:05:24,660
The core is where obligations live.
1790
01:05:24,660 –> 01:05:27,300
Compliance, finance, guardrails, SLA enforcement,
1791
01:05:27,300 –> 01:05:29,300
access boundaries, approval thresholds,
1792
01:05:29,300 –> 01:05:31,980
anything that can trigger an irreversible action,
1793
01:05:31,980 –> 01:05:35,380
create legal exposure or materially impact customers
1794
01:05:35,380 –> 01:05:36,900
gets deterministic logic.
1795
01:05:36,900 –> 01:05:40,700
Every time, same input, same output, same owner, same record,
1796
01:05:40,700 –> 01:05:42,460
because when a regulator and auditor
1797
01:05:42,460 –> 01:05:45,700
or an incident reviewer asks, why did the system do that?
1798
01:05:45,700 –> 01:05:47,500
You don’t get to answer because the model thought
1799
01:05:47,500 –> 01:05:50,740
it was a good idea where you need rule version, threshold,
1800
01:05:50,740 –> 01:05:54,580
evaluated value, action pathway, and who approved the exception.
1801
01:05:54,580 –> 01:05:55,540
That’s determinism.
1802
01:05:55,540 –> 01:05:57,820
AI does not belong in the rule engine.
1803
01:05:57,820 –> 01:05:59,220
It belongs in three places.
1804
01:05:59,220 –> 01:06:01,820
Explanation, summarization, and option generation.
1805
01:06:01,820 –> 01:06:03,740
Explanation means the system can translate
1806
01:06:03,740 –> 01:06:06,180
a deterministic event into human language.
1807
01:06:06,180 –> 01:06:08,540
Forecast variance trigger fired because segment A
1808
01:06:08,540 –> 01:06:11,100
stayed below threshold for 10 business days.
1809
01:06:11,100 –> 01:06:13,060
Primary drivers were pipeline velocity
1810
01:06:13,060 –> 01:06:14,740
and late stage deal aging.
1811
01:06:14,740 –> 01:06:16,060
That’s not the system deciding.
1812
01:06:16,060 –> 01:06:18,500
That’s the system reporting what it already decided.
1813
01:06:18,500 –> 01:06:21,140
In words, a leader can consume quickly.
1814
01:06:21,140 –> 01:06:23,180
Summarization means compressing state.
1815
01:06:23,180 –> 01:06:24,940
Three incidents are at breach risk.
1816
01:06:24,940 –> 01:06:27,100
Two are escalated to tier two.
1817
01:06:27,100 –> 01:06:28,580
One is overdue for acknowledgement.
1818
01:06:28,580 –> 01:06:30,740
The incident commander role is unassigned.
1819
01:06:30,740 –> 01:06:32,380
Again, not decision making.
1820
01:06:32,380 –> 01:06:35,780
Decision visibility option generation is where AI can be genuinely
1821
01:06:35,780 –> 01:06:38,660
useful because it can propose bounded interventions.
1822
01:06:38,660 –> 01:06:40,500
Based on previous variance recoveries,
1823
01:06:40,500 –> 01:06:42,100
here are three action pathways that
1824
01:06:42,100 –> 01:06:44,140
stayed within margin constraints.
1825
01:06:44,140 –> 01:06:46,020
Or based on similar incidents, here
1826
01:06:46,020 –> 01:06:49,140
are remediation steps that reduced time to mitigate.
1827
01:06:49,140 –> 01:06:51,860
But options are not actions and suggestions are not authority.
1828
01:06:51,860 –> 01:06:54,340
So the boundary is enforced with confidence thresholds
1829
01:06:54,340 –> 01:06:55,100
and stop rules.
1830
01:06:55,100 –> 01:06:58,060
If the AI has low confidence, it stops and escalates.
1831
01:06:58,060 –> 01:07:01,100
If the request implies an action outside the allowed playbook,
1832
01:07:01,100 –> 01:07:02,660
it stops and escalates.
1833
01:07:02,660 –> 01:07:05,540
And if the user asks for an exception that violates policy,
1834
01:07:05,540 –> 01:07:06,940
it stops and escalates.
1835
01:07:06,940 –> 01:07:09,620
This is the agentic autonomy curve in plain language.
1836
01:07:09,620 –> 01:07:12,340
The system can be creative only where creativity is safe
1837
01:07:12,340 –> 01:07:15,140
and it must be deterministic where consequences are expensive.
1838
01:07:15,140 –> 01:07:16,500
Now, the uncomfortable truth.
1839
01:07:16,500 –> 01:07:18,020
Most organizations do the opposite.
1840
01:07:18,020 –> 01:07:21,380
And they keep the core ambiguous, then ask AI to make sense of it.
1841
01:07:21,380 –> 01:07:22,860
So the agent starts stitching together
1842
01:07:22,860 –> 01:07:25,180
inconsistent definitions, stale refresh windows,
1843
01:07:25,180 –> 01:07:26,340
and missing state.
1844
01:07:26,340 –> 01:07:28,420
It produces an answer that sounds coherent
1845
01:07:28,420 –> 01:07:30,460
and leadership treats it like truth,
1846
01:07:30,460 –> 01:07:33,220
because humans confuse fluency with accuracy.
1847
01:07:33,220 –> 01:07:35,660
That is how you end up automating the wrong actions
1848
01:07:35,660 –> 01:07:38,140
for the right reasons on the wrong data.
1849
01:07:38,140 –> 01:07:40,540
And then the post-mortem says, AI failed.
1850
01:07:40,540 –> 01:07:42,420
When the real failure was architectural,
1851
01:07:42,420 –> 01:07:45,260
you never built a decision-grade substrate for the agent
1852
01:07:45,260 –> 01:07:46,260
to stand on.
1853
01:07:46,260 –> 01:07:47,500
So where is AI banned?
1854
01:07:47,500 –> 01:07:49,940
AI is banned from triggering escalations, freezing
1855
01:07:49,940 –> 01:07:52,220
spend, changing forecast baselines, assigning
1856
01:07:52,220 –> 01:07:54,860
severity, overriding ownership and marking actions
1857
01:07:54,860 –> 01:07:55,540
as complete.
1858
01:07:55,540 –> 01:07:57,500
Anything that changes operational state
1859
01:07:57,500 –> 01:07:59,860
must be backed by deterministic criteria
1860
01:07:59,860 –> 01:08:02,060
and recorded in the ledger with a reason.
1861
01:08:02,060 –> 01:08:03,700
The agent can initiate a workflow that
1862
01:08:03,700 –> 01:08:05,780
asks for human confirmation, but it can’t
1863
01:08:05,780 –> 01:08:09,540
unilaterally execute state changes that you’d regret in court.
1864
01:08:09,540 –> 01:08:11,300
And if you think that sounds restrictive, good.
1865
01:08:11,300 –> 01:08:13,660
Restriction is how control planes stay sane.
1866
01:08:13,660 –> 01:08:15,580
AI also doesn’t get to invent lineage.
1867
01:08:15,580 –> 01:08:18,300
It doesn’t get to blend sources because it feels helpful.
1868
01:08:18,300 –> 01:08:20,100
It can only cite certified data products
1869
01:08:20,100 –> 01:08:21,620
and approved semantic models.
1870
01:08:21,620 –> 01:08:24,420
If it can’t ground, it can’t speak with authority.
1871
01:08:24,420 –> 01:08:25,980
That distinction matters because this
1872
01:08:25,980 –> 01:08:27,620
is how you keep co-pilot everywhere
1873
01:08:27,620 –> 01:08:29,780
from becoming co-pilot made it up.
1874
01:08:29,780 –> 01:08:31,500
So the hybrid architecture is simple.
1875
01:08:31,500 –> 01:08:34,180
Deterministic decision rules define when something is true
1876
01:08:34,180 –> 01:08:36,300
and what must happen next.
1877
01:08:36,300 –> 01:08:39,060
Probabilistic AI helps humans understand, navigate,
1878
01:08:39,060 –> 01:08:41,180
and choose within that bounded space.
1879
01:08:41,180 –> 01:08:42,500
The system decides.
1880
01:08:42,500 –> 01:08:43,740
The agent explains.
1881
01:08:43,740 –> 01:08:46,900
The human overrides explicitly when required.
1882
01:08:46,900 –> 01:08:49,260
That’s how you get speed without losing governance.
1883
01:08:49,260 –> 01:08:51,340
And it’s how you make AI auditable.
1884
01:08:51,340 –> 01:08:53,900
By forcing it to operate on top of deterministic state,
1885
01:08:53,900 –> 01:08:55,940
instead of pretending it can replace it,
1886
01:08:55,940 –> 01:08:57,340
now that those boundaries are clear,
1887
01:08:57,340 –> 01:09:00,020
the final question becomes implementation posture,
1888
01:09:00,020 –> 01:09:02,180
not tool selection.
1889
01:09:02,180 –> 01:09:05,740
Operating principles leaders can actually enforce Monday morning.
1890
01:09:05,740 –> 01:09:07,620
Monday morning operating principles.
1891
01:09:07,620 –> 01:09:09,340
So what does Monday morning look like
1892
01:09:09,340 –> 01:09:11,580
when you stop pretending you’re building dashboards
1893
01:09:11,580 –> 01:09:13,900
and start admitting you’re building a control plane?
1894
01:09:13,900 –> 01:09:15,580
It starts smaller than you want and stricter
1895
01:09:15,580 –> 01:09:16,620
than you’re comfortable with.
1896
01:09:16,620 –> 01:09:18,420
Pick two decision surfaces exactly two.
1897
01:09:18,420 –> 01:09:20,140
One finance, one operations,
1898
01:09:20,140 –> 01:09:22,220
revenue forecast variance and incident.
1899
01:09:22,220 –> 01:09:25,540
SLA breach risk are good because they force honesty.
1900
01:09:25,540 –> 01:09:27,420
Money and outages don’t negotiate.
1901
01:09:27,420 –> 01:09:29,820
Now define the KPI as a rule, not a number.
1902
01:09:29,820 –> 01:09:31,820
If you remember nothing else, remember this.
1903
01:09:31,820 –> 01:09:34,500
A KPI is eligible for leadership attention only
1904
01:09:34,500 –> 01:09:35,820
when it carries an obligation.
1905
01:09:35,820 –> 01:09:38,620
So write the rule in a way the system can execute,
1906
01:09:38,620 –> 01:09:40,900
threshold, duration and scope.
1907
01:09:40,900 –> 01:09:43,940
Then attach the five non-negotiables you already heard.
1908
01:09:43,940 –> 01:09:48,020
Trigger, owner, action, time window, feedback.
1909
01:09:48,020 –> 01:09:50,780
That’s your definition of done, not a visual.
1910
01:09:50,780 –> 01:09:53,940
Next, enforce eligibility.
1911
01:09:53,940 –> 01:09:55,660
This is where most programs collapse
1912
01:09:55,660 –> 01:09:58,140
because they allow almost good enough to sneak in.
1913
01:09:58,140 –> 01:10:00,260
If the metric doesn’t come from a certified data set
1914
01:10:00,260 –> 01:10:01,820
with a refresh contract, you can defend.
1915
01:10:01,820 –> 01:10:03,060
It doesn’t get automation rights.
1916
01:10:03,060 –> 01:10:04,260
It can still be observed.
1917
01:10:04,260 –> 01:10:05,740
It can still be discussed.
1918
01:10:05,740 –> 01:10:08,940
But it cannot page humans free, spend or open bridges.
1919
01:10:08,940 –> 01:10:09,780
Same for logic.
1920
01:10:09,780 –> 01:10:12,220
If the definition isn’t centralized in a semantic model
1921
01:10:12,220 –> 01:10:13,540
with versioning and approval gates,
1922
01:10:13,540 –> 01:10:14,700
it doesn’t become a trigger.
1923
01:10:14,700 –> 01:10:17,220
No exceptions, the whole point is to stop building systems
1924
01:10:17,220 –> 01:10:18,860
that depend on tribal memory.
1925
01:10:18,860 –> 01:10:21,020
Then build the decision stack incrementally.
1926
01:10:21,020 –> 01:10:24,060
One layer at a time without trying to implement fabric
1927
01:10:24,060 –> 01:10:25,820
or roll out co-pilot.
1928
01:10:25,820 –> 01:10:28,180
Those are procurement words, not architectural ones.
1929
01:10:28,180 –> 01:10:29,180
Start with convergence.
1930
01:10:29,180 –> 01:10:31,540
Identify the minimum data set needed for the trigger
1931
01:10:31,540 –> 01:10:32,580
and certify it.
1932
01:10:32,580 –> 01:10:33,580
Don’t boil the ocean.
1933
01:10:33,580 –> 01:10:35,980
Boil the single pot that will burn you first.
1934
01:10:35,980 –> 01:10:37,420
Then compile meaning.
1935
01:10:37,420 –> 01:10:39,300
Create the measures that evaluate the trigger
1936
01:10:39,300 –> 01:10:41,460
and publish them as the canonical definition.
1937
01:10:41,460 –> 01:10:43,820
This is where you’ll be tempted to allow local variance.
1938
01:10:43,820 –> 01:10:44,900
Don’t.
1939
01:10:44,900 –> 01:10:47,780
Variants are entropy generators with a friendly face.
1940
01:10:47,780 –> 01:10:50,260
Then at state, create the decision ledger.
1941
01:10:50,260 –> 01:10:52,740
A table that stores trigger instances, rule version,
1942
01:10:52,740 –> 01:10:55,060
owner assignment, time stamps, status outcome.
1943
01:10:55,060 –> 01:10:57,700
If it can’t be queried later, it doesn’t exist.
1944
01:10:57,700 –> 01:10:59,260
The ledger is your immunity to,
1945
01:10:59,260 –> 01:11:01,180
I thought someone else handled it.
1946
01:11:01,180 –> 01:11:02,740
Then enforce action.
1947
01:11:02,740 –> 01:11:05,660
Why are the trigger to a flow that creates the ledger record
1948
01:11:05,660 –> 01:11:08,060
assigns the owner and starts the clock?
1949
01:11:08,060 –> 01:11:09,740
Execute a pre-committed pathway.
1950
01:11:09,740 –> 01:11:13,020
If you can’t pre-commit the action, you haven’t designed a KPI.
1951
01:11:13,020 –> 01:11:14,660
You’ve designed a topic for debate.
1952
01:11:14,660 –> 01:11:16,780
And this is the uncomfortable governance move.
1953
01:11:16,780 –> 01:11:19,300
Require exception handling to be explicit.
1954
01:11:19,300 –> 01:11:21,460
Exceptions are allowed, but they must be typed,
1955
01:11:21,460 –> 01:11:24,340
time bound and attributable to a human decision.
1956
01:11:24,340 –> 01:11:26,780
The system should never quietly not do the thing.
1957
01:11:26,780 –> 01:11:28,220
Finally, expose the interface.
1958
01:11:28,220 –> 01:11:30,660
Let leaders ask for posture, not charts.
1959
01:11:30,660 –> 01:11:31,380
But what fired?
1960
01:11:31,380 –> 01:11:32,140
What’s overdue?
1961
01:11:32,140 –> 01:11:33,340
What got overridden?
1962
01:11:33,340 –> 01:11:34,260
What worked?
1963
01:11:34,260 –> 01:11:36,700
And forced the interface to answer only from approved logic
1964
01:11:36,700 –> 01:11:37,500
and recorded state.
1965
01:11:37,500 –> 01:11:38,860
That’s how you keep the conversation
1966
01:11:38,860 –> 01:11:40,700
from becoming a narrative generator.
1967
01:11:40,700 –> 01:11:42,340
Now, how do you measure whether this worked?
1968
01:11:42,340 –> 01:11:44,700
You don’t measure it by how pretty the overview looks.
1969
01:11:44,700 –> 01:11:46,460
You measure it by decision latency.
1970
01:11:46,460 –> 01:11:48,500
Time from condition met to action started.
1971
01:11:48,500 –> 01:11:51,020
That’s the multiplier metric because it changes outcomes
1972
01:11:51,020 –> 01:11:52,620
across every domain.
1973
01:11:52,620 –> 01:11:54,780
In finance, you’ll see it immediately.
1974
01:11:54,780 –> 01:11:57,980
Interventions happen inside 48 hours instead of two weeks.
1975
01:11:57,980 –> 01:12:01,060
Forecast drift gets addressed while it’s still cheap to correct.
1976
01:12:01,060 –> 01:12:02,740
And because actions are recorded,
1977
01:12:02,740 –> 01:12:04,700
the organization stops relitigating
1978
01:12:04,700 –> 01:12:06,300
the same variance every month.
1979
01:12:06,300 –> 01:12:08,500
In IT, you measure breach prevention rate,
1980
01:12:08,500 –> 01:12:09,660
not just breach count.
1981
01:12:09,660 –> 01:12:11,860
If your system escalates before breach
1982
01:12:11,860 –> 01:12:13,220
and the playbooks launch on time,
1983
01:12:13,220 –> 01:12:15,020
you should see SLA compliance rise
1984
01:12:15,020 –> 01:12:17,820
because you stop discovering urgency by accident.
1985
01:12:17,820 –> 01:12:19,660
This is also where you measure entropy directly,
1986
01:12:19,660 –> 01:12:21,420
even if you don’t call it entropy.
1987
01:12:21,420 –> 01:12:24,700
Track override rates, track false trigger rates,
1988
01:12:24,700 –> 01:12:26,260
track how often ownership resolves
1989
01:12:26,260 –> 01:12:28,780
cleanly versus how often it bounces.
1990
01:12:28,780 –> 01:12:31,300
Those are signals that your decision engine is drifting
1991
01:12:31,300 –> 01:12:32,580
and don’t miss the leadership point.
1992
01:12:32,580 –> 01:12:33,980
This is not an IT project.
1993
01:12:33,980 –> 01:12:36,020
This is organizational design with software.
1994
01:12:36,020 –> 01:12:37,820
You’re deciding where meaning leaves,
1995
01:12:37,820 –> 01:12:40,540
who is obligated, what actions are permitted
1996
01:12:40,540 –> 01:12:42,460
and which exceptions are tolerable.
1997
01:12:42,460 –> 01:12:45,220
Fabric semantic models, data verse, flows, agents,
1998
01:12:45,220 –> 01:12:47,340
those are just the implementation substrate
1999
01:12:47,340 –> 01:12:50,180
for a decision contract you either enforce or you don’t.
2000
01:12:50,180 –> 01:12:52,420
Monday morning, you’re not asking for more KPIs.
2001
01:12:52,420 –> 01:12:55,780
You’re attaching obligations to the few that actually matter.
2002
01:12:55,780 –> 01:12:57,780
Stop asking for one page KPIs
2003
01:12:57,780 –> 01:13:00,460
and start demanding deterministic decision surfaces,
2004
01:13:00,460 –> 01:13:02,540
triggers, owners, actions, time windows
2005
01:13:02,540 –> 01:13:05,420
and a ledger that proves the organization responded.
2006
01:13:05,420 –> 01:13:08,220
If this episode made you rethink how your org runs
2007
01:13:08,220 –> 01:13:11,500
on dashboards, leave a review based on whether it clarified
2008
01:13:11,500 –> 01:13:13,700
what decisions should be engineered next.
2009
01:13:13,700 –> 01:13:15,540
And if you want to argue with me in public,
2010
01:13:15,540 –> 01:13:17,740
connect with me on LinkedIn, Miracol Peters
2011
01:13:17,740 –> 01:13:20,500
and send me your worst one page KPIs request.
2012
01:13:20,500 –> 01:13:22,740
Tell me which decision surface you want dissected next
2013
01:13:22,740 –> 01:13:24,580
and I’ll pull it apart.
