How I write the decision

Three documents from work already on this site. No invented numbers. Where we did not measure, the doc says what I would instrument now.

One-pager

Rules wizard

Problem: New MSP admins stalled on an empty Rules page. The setup path was company, then Microsoft enterprise apps, then policies.
Hypothesis: A guided first-run wizard cuts time-to-complete for the first successful setup, including tenants whose policies already conflict.
Scope: The first-run path only. A prototype, including an AI pass for the draft. Not a rewrite of the rules editor.
How I would measure: Time-to-first-successful-policy, and drop-off on each step. A win: the admin finishes setup without a “where do I start” ticket. We did not instrument this.
Not doing: Rules that apply themselves. A wizard on every dense page. The pattern shipped on Rules and stopped there.

Kill memo

Bet: Show idle time in dollars for a client manager in a recession, so the weekly conversation has a number the buyer already speaks.
How I would know: About two managers per company open the figure and use it in a conversation. If they do not, the report is decoration.
Cost of the test: It needed input from the client and a CRM integration. That is a build, not a prototype.
Decision: I killed it. The test cost more than the learning. Right kill.

Metric tree

North star I would name: There was no formal north star on the team that I owned. The bar I actually used: an admin finishes the task in a short time-to-complete.
Proxy: Time-to-first-successful-policy for the wizard. Whether the weekly PDF is used in an account conversation. Time-to-first-valid-rule against the click editor.
Guardrail: A new feature must not break settings already coupled to it. Watch support tickets about broken access after a release.
Honest limit: This tree was not on a dashboard. It is the one I would put up now, from the same work.