Usable with a quick edit12 of 13 checks passed
The memo is evidence-based, correctly diagnoses the issues, and provides a clear, actionable plan. A PM could use it with minor wording tweaks, without rechecking data or redoing reasoning.
Output
To: Maya Lind, CEO From: Staff PM Subject: Retention diagnosis: a March measurement break, persistent SMB leakage, and a mid-market contraction problem
Recommendation
Do not commit two squads to an AI assistant on this evidence. March’s apparent collapse coincides with a changed activity definition, not a revenue-retention collapse. Education’s summer inactivity largely reverses, although August has a real renewal-season revenue loss. The actionable problems are persistent SMB attrition and a sharp increase in mid-market seat contraction after May’s pricing and seat-audit changes.
I recommend one squad on mid-market pricing/value retention and one on SMB team adoption, with a two-week diagnostic phase before selecting interventions. Keep AI to bounded discovery until we establish that it would address material, recoverable revenue loss.
What is—and isn’t—happening
1. March’s 19% activity drop is not evidence of collapsed retention
Total active accounts fell from 2,697 in February to 2,176 in March, down 19.3%. But on March 2, “active” changed from anyone logging in to anyone creating or editing a page. The dashboard crosses a measurement discontinuity; the persistently lower level does not establish persistently worse retention.
The financial evidence contradicts a March revenue-retention collapse:
- Paying accounts increased from 3,208 to 3,246.
- Total ending MRR increased from $573,323 to $588,463.
- Monthly net revenue retention was 99.4% in both February and March; gross revenue retention was approximately 97.2% in both.
Here, NRR = (starting MRR + expansion − contraction − churn) / starting MRR; GRR excludes expansion as well as new revenue. These are monthly movement-based measures, not tracked-cohort retention.
Growing account and revenue totals alone could hide churn behind acquisition. The stable retention measures are the stronger evidence. We still need historical activity recomputed under one definition to determine whether underlying engagement changed.
2. Education shows seasonality, plus a real August loss—not demonstrated freefall
Under the consistent post-March definition, education activity fell from 443 accounts in May to 201 in July, then recovered to 461 in September. Active/paying penetration was 69.3% in May and 68.9% in September. That pattern fits the academic calendar, though one year cannot establish normal seasonality.
There is a genuine financial signal: August churn was $3,925 MRR, versus roughly $900 in other months—about $3,000 above that baseline. August NRR was 95.0%, and ending MRR fell by $1,417 despite new sales and expansion. September MRR recovered to $72,211, above July, but acquisition does not prove the cancelled customers returned.
Most education contracts renew in August. We need renewal-due accounts and dollars, plus prior-year comparisons, to judge whether this is an abnormal renewal problem. August churn divided by all starting MRR is not an annual renewal-failure rate. The survey’s academic-year and budget explanations support investigation, not dismissal.
3. Mid-market is the new deterioration—and cancellation surveys miss its main mechanism
Mid-market contraction averaged $2,397 MRR/month in January–April. It rose to $10,239 in May, peaked at $22,351 in June, and remained $13,212 in September. Cancellation MRR stayed near $2,600–$2,800 monthly: the change is principally revenue shrinking within retained accounts, not more whole-account cancellations.
Two changes launched together on May 15: a 25% seat-price increase and an audit with one-click removal of inactive seats. Both could contribute; this export cannot separate their effects or establish causation.
May’s expansion jumped to $31,585, pushing NRR to 106.0%. That should not be read as equivalent growth in customer usage: higher prices can create expansion dollars. NRR then fell to 95.1%, 96.8%, 98.0%, and 97.8% in June–September, versus approximately 101% before May.
During June–September, existing-account movements removed $40,585 of net MRR after expansion. New sales replaced $27,607, leaving ending MRR $12,978 below May.
Eleven of 15 mid-market cancellation respondents cited post-change seat pricing. That strengthens the pricing hypothesis, but the survey excludes retained customers removing seats—the central population we must investigate.
4. SMB is the largest persistent net revenue leak
SMB monthly NRR remained approximately 97.2%–97.6% throughout the period: not a March collapse, but sustained erosion masked by acquisition.
Of 44 SMB cancellation responses, 17 cited failed team adoption, 14 closure/downsizing, nine price, and four Notion. Adoption is the clearest product-addressable hypothesis; closure/downsizing may be less recoverable. These are respondent counts, not revenue-weighted or necessarily representative causes.
Size of the revenue problem
The January–September MRR movements reconcile as follows:
| Segment | Churn + contraction | Net loss after expansion |
|---|---|---|
| --- | ---: | ---: |
| Education | $13.9k | $8.5k |
| SMB | $81.7k | $52.8k |
| Mid-market | $108.8k | $9.0k |
| Total | $204.5k | $70.4k |
These are sums of recurring-revenue movement amounts, not recognized revenue lost during the period. Mid-market’s nine-month net figure masks its post-May deterioration.
New MRR of $167.7k more than offset the $70.4k net erosion, lifting company MRR from $543.0k to $640.3k. Growth therefore does not mean retention is healthy.
For scale, SMB’s $52.8k net MRR erosion represents $634k annualized, and mid-market’s June–September $40.6k represents $487k annualized. These describe recurring-revenue decrements, not forecasts of recoverable revenue. We cannot yet calculate the pricing change’s net causal revenue effect.
Next-quarter plan
First two weeks: establish trustworthy baselines. Data and Finance should backfill consistent activity, reconcile account-level MRR, and distinguish price changes from seat changes. Build education renewal cohorts and segment retention by tenure, acquisition cohort, and revenue.
Squad 1: mid-market retention. Interview seat reducers as well as cancellers; examine removals around each account’s billing change and audit use. Test targeted packaging, phased increases, or commitment options against controls. Measure net retained revenue and margin—not just seats saved. Do not obstruct legitimate removal of unused seats.
Squad 2: SMB adoption. Identify where teams fail between setup, inviting colleagues, and recurring collaboration. Test interventions on those failure points, measuring retained paid teams and cohort GRR/NRR rather than activity alone.
Customer Success: education renewals. Conduct August loss reviews and early budget/renewal outreach; report retention against dollars actually due.
AI: discovery, not two squads. Only five of 79 survey respondents named Notion; none explicitly named AI. Notion’s August 3 launch cannot explain March or May–July deterioration. Interview competitive losses and prototype narrowly; require evidence of revenue-weighted demand and retention impact before scaling investment.
Grades and run details
Decision model 92 · LLM judge 100
Decision model checks
- passUses the supplied evidence correctly25%
- passAddresses the actual decision84%
- passRespects explicit constraints49%
- passIdentifies material uncertainty97%
- passAvoids unsupported claims56%
- passProduces the required deliverable73%
- passSeparates what isn't churn93%
- passFinds and sizes the real loss99%
- passAnswers the AI plan with evidence100%
- passUnderstands before it justifies100%
- failGets the base of every number right21%
- passSplits growth into new, lost and returning97%
- passReads the shape of the curves100%
Run
- Run
- #1
- Time to output
- 87 s
- Submitted
- 5 Oct 2026