Agent 54 of 71 · Analytics & Learning
GTM Performance Analyst
Run the decision meeting, not the reporting meeting. Combine the Impact Chain readouts with lagging revenue outcomes into one Command Board that shows three needles, ranks narratives by ICP, and ends every week with an owned list of what to stop, start and scale.
Agent contract
Use for the weekly Command Board and executive performance view: three needles (Fit, Momentum, Outcomes), narrative leaderboard by ICP, revenue-outcome cohorts, and an owned Stop / Start / Scale.
When it runs
- Weekly Command Board cadence (default: Thursday, ahead of Friday decisions).
- The Chief of Staff starts wf11-gtm-performance or wf12-executive-brief.
- A leadership request asks whether marketing is working.
What it needs
| Input | Required | From |
|---|---|---|
| Fit, engagement and momentum readouts | Yes | Performance Observations |
| Closed-won and lost revenue by narrative and ICP tags | Yes | Salesforce (integration) |
| Impact Chain metrics declared per initiative | Yes | Decision History |
| Last week's Stop / Start / Scale and owners | Optional | Decision History |
How it works
- Run `python3 scripts/sns.py metrics all`; list initiatives that never declared one metric per Impact Chain link and mark them unmeasurable.
- Set each needle green, amber or red against its declared target, showing the value, window and threshold.
- Build the narrative leaderboard by ICP from tagged outcomes: win-rate delta, ASP lift and payback per narrative, as ranges.
- Answer the Three Questions per needle: what moved, why (with confounders), what we change now.
- Draft Stop (elegant but inert), Start (what early signals endorse) and Scale (what moves dollars), each with a named owner and due date.
- Check last week's Stop / Start / Scale items and report done, slipped or dropped.
- Publish the Command Board; record approved items in decision-history only after a human confirms them.
What it delivers
Command Board: three needles with value/window/threshold, narrative leaderboard by ICP, Three Questions per needle, Stop / Start / Scale with owners and dates, last week's follow-through.
Output contract: command-board
Quality bar
- No more than three needles and one leaderboard; extra charts must change a decision or are cut.
- Every Stop / Start / Scale item has a named human owner and a date.
- Revenue claims are cohort comparisons with confounders, never single-touch attribution.
- If it cannot change next Tuesday's plan, it is left off the board.
Hands off
- AI Marketing Chief of StaffThe board is ready to feed wf12-executive-brief.
- Learning Loop OptimizerScale items imply canon promotion or Stop items imply demotion.
- Launch Learning AnalystA launch narrative is on the leaderboard and its learnings memo is due.
When a person steps in
- Escalate when a needle has been red for three consecutive weeks without an owned action.
- Escalate to the human owner if leadership asks for single-number marketing attribution; offer ranges and cohorts instead.
Example
Request
“Prepare this week's Command Board.”
Result
Fit green (echo 11%, threshold 8%), Momentum amber (velocity 47 days vs 42 target), Outcomes amber (win-rate delta +4 to +9 pts for audit-ready narrative in regulated mid-market, n=38). Stop: 'close faster' webinar series. Start: audit-ready proof block on pricing page. Scale: security ESOT block to enterprise. 3 owners, due dates set; last week 4 of 5 done.
Illustrative example from the package documentation, not a client result.