# Trial conversion by cohort

"We converted 18 trials this month and started 60, so our conversion rate is 30 %." The 18
and the 60 are different people. Most of this month's conversions come from last month's
trials, and most of this month's trials have not decided anything yet. Divide them and you
get a number that goes up when signups fall and down when a campaign works.

## A cohort is everyone whose trial started in a month

Put every trial into the month it started, and ask what became of it, whenever that happened:

| Trials started in | Started | Paid at some point | Still in trial | Did not convert | Rate |
|---|---|---|---|---|---|
| June | 50 | 16 | 0 | 34 | 32 % |
| July | 64 | 19 | 0 | 45 | 30 % |
| August | 58 | 12 | 9 | 37 | 24 % of the decided |

A payment in July by somebody whose trial started in June belongs to June's row. June's
rate can still rise for a while (somebody comes back after two months), and that is fine: the
row tells the truth about June's trials at any time.

## Leave out what is not decided

August looks worse than July, but nine trials are still running. The honest rate divides by
the trials that are **decided**: converted, or ended without a payment. 12 of 49 is 24 %, and
it will move as the nine decide. Showing 12 of 58 (21 %) punishes every recent month and makes
every chart slope down at the right edge, which people then learn to ignore, together with
the real drops.

## When is a trial converted

At the first real payment, not at the end of the trial period. A trial that ends without a
payment method goes to "paused" in Stripe and never produces an invoice; a trial with a card
that fails is not a customer yet either. And the other direction: somebody who pays on day
three of a fourteen-day trial converted on day three.

That gives a second number worth as much as the rate: **days from trial start to first
payment.** A median of 15 days on a 14-day trial means people wait for the end. A median of 4
means they decide early and the rest of the trial is waiting time; a shorter trial or an
earlier offer may cost you nothing.

## Trials that never touch Stripe

If your product runs the trial itself and only creates the Stripe subscription at purchase,
Stripe knows nothing about the trials that did not convert: from Stripe's data your
conversion rate is 100 %. The trial has to come from where it lives. In sumwerk the product
sends the trial's start and end with the customer through the API; customers who only exist in
Stripe keep their trial dates from the trialing subscription. Both land in the same cohorts.

## What to look at each month

1. The rate of the last fully decided cohort, against the ones before it.
2. The median days to first payment, and whether it moved.
3. The size of the cohort: a rate on eight trials is noise.
4. The same by segment (plan, source, company type) once the cohorts are large enough: an
   average of 28 % is often 45 % in one group and 12 % in another, and those need different
   things from you.