Do teams keep paying for AI meeting notes after month three?
Answer
Not by default. Month-six retention for AI note-takers sits between 18% and 34% in every cohort we could read — unless the notes are written somewhere the team already works. Where a destination is connected in the first fortnight, month-six retention is two to three times higher in all four datasets that report it.
The lever is onboarding, not summarisation quality: no source shows summary accuracy predicting renewal once it clears roughly 90%.
Evidence
-
01
SecondaryBracken Research, State of Meeting Intelligence 2026, June 2026 · n=6 vendor cohorts ·
Month-six retention clusters in the low twenties
Across six vendors that disclose cohort curves, median month-six retention is 24%; best 34%, worst 18%. Month three is where the curve bends, not month one.
bracken-research.com/smi-2026 -
02
PrimaryInternal cohort analysis, 4,182 workspaces activated Jan–Mar 2026 · queried 11 Sep 2026
A connected destination is the single strongest correlate
Workspaces that connected a CRM, tracker or wiki within fourteen days retained at 71% by month six, against 22% for those that never did — a gap that holds after controlling for seat count and plan.
-
03
PrimaryBuyer interviews, 22 accounts, 20–120 seats, Jul–Aug 2026 · transcripts on file
Summary quality is table stakes, not a differentiator
Buyers rank accuracy in the top three reasons to buy and outside the top five reasons to renew. Every product in the set was "good enough" on transcription.
-
04
PrimaryInternal billing extract, FY2025–26 · plus two vendor disclosures reporting the same shape
Churn happens at the renewal date, not during the trial
Trial-to-paid is stable at 38%, but 81% of lost revenue leaves at the first annual renewal — which is why quarterly dashboards read this market as healthier than it is.
-
05
PrimaryExit interviews, 22 accounts · coded independently by two analysts, agreement 0.84
Price is not the stated reason — dormant seats are
Three of twenty-two churned accounts named price. Fourteen named seats they paid for and never used — the same complaint under another name.
Sources · 9
- 01Bracken Research, State of Meeting Intelligence 2026Jun 2026 · secondary
- 02Internal cohort analysis, 4,182 workspacesSep 2026 · primary
- 03Buyer interviews, 22 accounts, 20–120 seatsJul–Aug 2026 · primary
- 04–05Internal billing extract FY2025–26; exit interviews, 22 accounts, double-codedAug–Sep 2026 · primary
- 06–09Four vendor investor updates and pricing pages, read 8–10 Sep2026 · secondary
What would change this answer
- An audited cohort curve. Two of six are self-reported with no methodology note; if either is out by five points the spread widens enough to matter.
- A destination-connection experiment. The 71%/22% split is correlational; a forced-connection onboarding test settles it in one quarter.
- A second renewal cycle. Every dataset stops at month six or the first renewal. Nothing here says what month eighteen looks like.
Still unknown
- Whether the destination effect survives when the destination is a wiki nobody reads.
- How much of the month-three bend is meeting volume falling, not the tool failing.
- Whether teams under twenty seats behave like the mid-market at all: 61 in the base, an interval too wide to quote.