Indexxero

Can your churn data actually tell you anything?

Most retention analysis is built on too few events to be conclusive, and nobody says so. Check yours in 30 seconds. No signup, no sales call, no email required.

Why 100 events, and not 20

Twenty events is enough for a single pooled estimate. A useful retention report is not one estimate. It splits churn three ways (voluntary, payment failure, contraction) and crosses that with three lead-time windows, plus a random control. That is nine to twelve separate cells, each needing roughly 20 to 30 events before the number means anything.

Below that, the honest answer is "inconclusive", and a report that says otherwise is selling you confidence it does not have.

Your shape	Events in 24 months	Verdict
4,000 accounts, 12% churn	960	Conclusive, with room to segment
1,200 accounts, 9% churn	216	Conclusive
600 accounts, 8% churn	96	Borderline, headline only
200 accounts, 6% churn	24	Not conclusive

What we do when the data clears the bar

We rebuild your last 24 months exactly as they looked at each past date, not as they look now, and produce a board document that answers three questions your dashboard cannot.

• What did you actually lose? ARR split into voluntary cancellation, payment failure and contraction. Most companies have never separated these, and each has a different fix.
• What was knowable in advance? How much churned ARR was already flaggable at 30, 60 and 90 days out, with a random baseline printed beside every figure so you can see whether the signal is real.
• How fast did you react? Days between the first recorded intervention and the cancellation, and how much ARR churned with no intervention at all.

You get a sealed forward forecast with it: our prediction for your next two quarters, hash committed before the period starts and scored publicly when it ends, whether or not we were right.

Terms

$15,000 fixed, ten business days, from CSV exports. No production access and no integration project. If your data cannot support conclusions we tell you within 48 hours and refund in full. We would rather refund than publish a number too wide to act on.

Two sample reports

The report adapts to how you bill. If you charge a card automatically, payment-failure churn is a real and separately measurable category. If you invoice against annual renewals, it structurally cannot happen, and we report it as not measurable rather than as zero. Pick whichever matches you.

Sample: card on file, monthly Sample: invoiced, annual

Send us your numbers

What this will not tell you

It will not prove that intervening on a flagged account would have saved it. That requires a randomised holdout running forward, where a share of at-risk accounts is deliberately withheld from outreach and assigned before outcomes are known. Everything in the backtest is prediction and correlation, and the report says so on its own page rather than burying it.

We say this up front because the alternative is a category where every vendor claims outcomes and none of them publish a hit rate.