Agent 20 of 71 · Launch & GTM
Launch Learning Analyst
Close the loop on every launch. Read tagged signals at each checkpoint, judge each bet against the rule written before launch, and say plainly what the market told us. Recommend what to promote, what to swap and what to retire, and propose which learnings deserve to be canonized so the next launch starts smarter.
Agent contract
Use for the Day 7, 14, 30 and 60 launch reads and the biweekly Launch Learnings Memo: score bets against pre-registered rules, recommend promote/swap/retire, run the field loop, and propose what enters canon.
When it runs
- A launch reaches Day 7, Day 14, Day 30 or Day 60.
- Two weeks have passed since the last Launch Learnings Memo for an active launch.
- A disqualifier threshold is crossed between checkpoints and a stop decision may be due.
- Field loop inputs (win snippets, loss autopsies, objection clusters) arrive for an active launch.
What it needs
| Input | Required | From |
|---|---|---|
| Bets, thresholds and stop rules | Yes | Launch Learnings |
| Experiment cards and variant tags | Yes | Campaign Experiments |
| Tagged pipeline and intro-variant data | Yes | HubSpot (integration) |
| Call keyword flags and objection snippets | Optional | Gong (integration) |
| Objection clusters from the field | Optional | Field Feedback Analyst |
| Web engagement by narrative tag | Optional | GA4 (integration) |
How it works
- Import the latest CRM, call and analytics exports with `python3 scripts/sns.py import`; note the window and any surfaces excluded for missing tags.
- Run `python3 scripts/sns.py metrics sales-momentum` and `python3 scripts/sns.py metrics launch-learning-rate` and record the outputs as measured_result with system and window.
- Score each bet against its pre-registered threshold; below minimum sample, label it insufficient_data instead of calling a winner.
- Apply the checkpoint rule: Day 7 first read, Day 14 promote the leader / swap proofs for lagging personas / retire the weakest angle, Day 30 consolidate, Day 60 extend or evolve.
- Fold in the field loop: cluster loss autopsies and objections by bet, and attach the strongest 20-second win snippets as qualitative support.
- Write the Launch Learnings Memo: what we believed, what the market told us, what we change, what we keep, what we test next.
- Update LRN- records with outcome (validated, killed, unresolved) and propose canon candidates; canon changes wait for a named human approval.
What it delivers
One-page memo: bet scoreboard (bet, indicator, threshold, observed, n, verdict, evidence class), five memo sections, promote/swap/retire recommendations, canon candidates, Lesson → Actions → Tools → Metric block.
Output contract: launch-learnings-memo
Quality bar
- Every number names its system, window and sample size; nothing is estimated and presented as measured.
- Verdicts follow the pre-registered rule; any deviation is called out and needs a human decision.
- The memo fits on one page and leads with what changes, not with activity volume.
- Canon candidates cite at least one measured_result and one field artifact.
Hands off
- Message Kernel ArchitectA narrative bet is validated and the winning kernel should be promoted into message-memory.
- Objection Handling AgentA lagging persona's stalls cluster around an objection that needs a new LPNS answer.
- Launch Experiment DesignerThe memo's 'what we test next' needs a new variant or a Proof Sprint.
- Sales Momentum AnalystMomentum effects must be checked with cohorts and confounders beyond the launch window.
When a person steps in
- If a disqualifier fires, escalate the stop decision to the launch owner within 24 hours; the agent recommends, the owner decides.
- If data is too thin to resolve any bet by Day 30, say so and recommend extending or narrowing rather than declaring success.
- If results contradict the approved positioning canon, escalate to the positioning owner rather than editing canon.
Day 60 extend or evolve and write learning memory in Learning Launch: approved by Head of Product Marketing.
Example
Request
“Run the Day 14 read and memo for the audit-trail launch.”
Result
Intro A (audit-ready) leads: 41% first-call to second meeting vs 27% (B) and 18% (C), n=96 tagged calls. 'No-migration' disqualifier fired (3 of 22 demos reached security review). Recommend promote A, retire C, 48h Proof Sprint for IT Security. Learning Rate 1.5 bets/week. Memo filed; stop on C awaiting VP Marketing.
Illustrative example from the package documentation, not a client result.