Working · 2025-09SecondaryRetentionAnalytics

Rebase retention to month 3 for AI-era products

For self-serve, month-to-month AI products, measure retention from month 3 onward instead of month 0, because 'AI tourist' signups churn out early.

Also called: M12/M3 ratio

Claim and evidence

The claimM3 represents your 'true' customer base after AI tourists have churned out.

Leading AI companies (a16z dataset) · a16z.com · 2025-09-10

What the numbers show

No published number for this tactic has passed our word-for-word check yet, so there is no chart.

Why it works

Diagram, not data
Cohort revenueby signup monthTourists churnin early monthsRebase to monththreeTrack latermonths againstit
Cohort revenue by signup monthTourists churn in early monthsRebase to month threeTrack later months against it

Early curves reflect hobby trial, not real demand. Month 3 is when most curves flatten, so M12/M3 predicts long-term net dollar retention better.

How to run it

  1. Cohort revenue by signup month
  2. Rebase to M3 as true customer base
  3. Track M12 divided by M3
  4. Watch for a rising 'smile' curve

Ways to run it

Design options to test, not promised results.

  1. Option AMonth-three baseline

    Treat month three as the true customer base and report later revenue as a share of it.

  2. Option BLate-to-early ratio

    Track the twelfth month against the third as the leading signal for long-term net dollar retention.

  3. Option CWatch for the smile

    Look for cohorts whose revenue rises again after the early drop, a sign of expansion in the core base.

Dated examples

  1. 2025

    Leading AI companies (a16z dataset)

    Dozens of AI companies above $1M ARR analyzed by cohort.

    “M3 represents your 'true' customer base after AI tourists have churned out.”
    ✓ quote checked word-for-word on the page

    a16z.com · 2025-09-10

Does it still work?

Workingas of 2025-09

Sept 2025 essay; sample undisclosed and charts not readable in the fetched text. source

Risks

  • Undisclosed sample
  • Not validated for non-AI SaaS

Sources

  1. a16z.com · a16z · Santiago Rodriguez and Alex Immerman · 2025-09-10
    “Most often, the curve begins to flatten around M3.”
    ✓ quote checked word-for-word on the page

Change log

  • 2026-11-20 — harvest v0.1 (retention)