Estimating a Dynamics 365 implementation is a lot like looking into the future with limited visibility and knowledge, however many times there is better information but it needs to be gathered. This part can take a lot of time.
So I built a skill that runs the interview. It only works in Cowork, and the install is three clicks: Customize, then Skills, then Add and upload the zip built from the repo. After that it shows up under the Skills tab when you start a new task.
What it does
It asks the questions one at a time and writes your answers into a JSON data file. You don’t have to prep anything first.
In the video I walk through a demo for a fictional manufacturer: an 8 month finance-only first wave, 400 development hours, 30 Finance Premium and 20 Finance base licenses purchased up front, one payroll integration, no ISVs, Copilot for Finance with M365 Copilot across the finance org, one production environment, 36 month projection. That file loads into the estimator page and populates the entire model month by month at Microsoft list price.
Answer to the best of your ability and let the estimator’s experience fill the gaps. When I said one developer for 400 hours, the follow-up was that one is too few for an eight month timeline, so we went with two. That is exactly the interaction you want. A salesperson who doesn’t know the right number still produces a complete, reviewable input set instead of a blank page.
What it does not do
It does not produce a number you can rely on immediately, i.e. share with your boss or send to a client.
List prices are not real prices, it is a starting place for a negotiation, hence the estimator tool has a place for you to put the actual negotiated price.
As an experienced Dynamics 365 consultant, I knew Cowork and the agents starting in month eight, which I want to move. I also had to strip out environments I did not need and add a month for the demo environment. Many times I’ll shape the development environments to better match when design will be done ‘enough’ to begin development but early enough to have extensions and interfaces ready for testing and refinement.
This wasn’t an AI defect or hallucination. The estimate is a starting draft, and every one of those adjustments took seconds because the structure was already there.
Key Point
This AI automation did not replace an expert who owns the final estimate and makes judgment calls based on experience. How many developers for 400 hours. Whether the payroll interface is a real integration or an Excel import in disguise. Whether Fabric belongs in the number at all.
Thankfully, the person collecting the inputs no longer has to be the person who knows the answers. Sales gathers, the architect judges, and the two are no longer waiting on each other’s calendar; the handoff can happen asynchronously.
That is the true value here- the reduction in overall back and forth and process time which us lean experts call WASTE!
The code is MIT-licensed on GitHub. Issues and pull requests welcome. If you have ideas, let me know!
Let me know what you think!
you might also like