Skip to main content

​
Video walkthrough

At the org level the questions get bigger. Where is activity dropping off. Which groups have stopped practicing. Which skills are weak across every team. Where the training investment should go next quarter. Those are hard questions to answer in a dashboard, because you end up pulling one report at a time and stitching the picture together yourself. Asked through a connected assistant, the stitching is the part that goes away. This page needs workspace admin, which is the same permission that governs analytics inside Exec. A group admin can run all of it narrowed to their own groups.
Not connected yet? See Connect your AI tool. If you have several tools connected, open with “Using the Exec MCP, …” so your assistant knows which one you mean.

​
1. See where practice is actually happening

What comes back is volume and its trend, then the same split per group, so you can see which teams carry the activity and which have gone quiet. Read it for the shape rather than the total:
  • Groups that have fallen off, which is the earliest signal you get that a rollout is not sticking.
  • Groups where low volume is expected. Not every group should be practicing at the same rate, and a team of program leads reading low is not a problem. Knowing which is which is the judgment the dashboard cannot make for you.
  • Momentum, not just volume. A group that has doubled off a small base and a group that is flat off a large one need different conversations.
Then go after whatever looks wrong:
Exec hands your assistant one group at a time rather than a ready-made comparison table, so the assistant pulls each group and assembles the breakout itself. Two consequences: your groups have to exist in Exec for any of this to work, and on a large workspace naming the groups you actually care about is quicker than asking for all of them.

​
2. Map skill strength across every team

The sample-size clause is not padding, and at org scale it matters more than anywhere else. A workspace can carry dozens of skills, and the ones with the most extreme scores are almost always the ones almost nobody has been scored on. Without the clause your weakest skill is usually just your least-practiced skill, and your strongest is a skill with one good session behind it. Set the floor to something defensible for your headcount and say it out loud in the prompt. Ask for the real thing is the five minutes of reading that keeps a number like this out of a board deck before it has been sanity-checked. What comes back is a grid of skill against group, with the absolute score, the workspace average, and the spread between groups. The spread is the column worth reading first: a skill everybody is mediocre at is a content problem, and a skill one group is strong at and another is weak at is a coaching or a hiring problem.

​
3. Check it against live calls

If you have call scoring set up, the same conversation can ask what is happening on real calls, not just in practice.
You get the spread of scored calls and then the breakdown by category, which is usually where a clear laggard shows up. One category averaging a rank below the others is a more useful finding than any single call. Then ask why:

​
4. Find the coverage gap

You get completion, not passed, not started, and past due, first across the workspace and then per group. The best and worst groups are the obvious read, and they are worth having. The finding that usually matters more is the one nobody asks for: the groups with nothing assigned at all. Completion rates only describe people who were given something. On most workspaces a meaningful share of seats have never been assigned anything, and they are invisible in every completion number because they are not in the denominator. Ask for it directly:
A low completion rate is a follow-up problem. A group with zero assignments is a program design problem, and it is usually the larger number.

​
5. Turn it into something you can hand over

Everything above is in one conversation, so the summary does not need restating.
The first pass usually comes back thorough rather than short. That is worth saying out loud:
What you get is a document you can drop into a board deck, a QBR, or the next enablement meeting. The writing is the part that saves the most time, and it is the part an export cannot do for you.
Before this reaches a board, check what each headline figure is based on. Ask what is that number based on? and you get the sample behind it. The same diligence you would apply to any number you present.

​
Set up a project so you do not repeat yourself

This is the habit that separates trying MCP once from using it every quarter. Most assistants let you save a workspace with its own standing instructions. Claude calls these projects, ChatGPT has custom instructions, and other tools have their own version. Put Exec in one and write down what you would otherwise retype:
  • Your group structure and how you want it broken out
  • Your standing sample-size floor, so you never have to remember the clause again
  • The house rules for numbers, such as always saying what a claim is based on and only counting ranked sessions
  • What to leave out, which matters when you are going to share your screen
After that, “give me the quarterly readout” is enough, because the standing instructions carry the rest.

​
Make it your own

These five are a starting point. The questions that end up mattering are the ones specific to how your organization is structured, so it is worth asking for things this page does not mention and seeing how far it goes. If something does not work the way you expect, Troubleshooting covers the predictable failures, and [email protected] will pick up anything it does not.

​
Next

Prompts and reliable answers

The full prompt library, and how to check a number before it reaches a board deck

Managing your team

The same walkthrough one level down, for the managers you are rolling this out to

What the AI can see

The permissions model, in a form you can forward to a security reviewer