The trial-to-paid number most indies misread

A developer wrote to us in August, worried. Trial-to-paid for a sleep app had fallen from 44% to 29% in a month. Nothing had changed in the app. Had something broken?

Nothing had. The app had been featured, three times as many people had started a trial, and the new arrivals were less sure about paying than the regulars. Paying subscribers were up 96%. The percentage had fallen because the business had grown.

Trial-to-paid is the number subscription developers quote most and, in our inbox at least, the one most often read backwards. Here are the three ways it misleads.

It is measured on the wrong day

A seven-day trial that starts on the 28th converts, or does not, on the 5th of the next month. If you divide this month’s conversions by this month’s trial starts, you are comparing two different groups of people. In a flat month the error is small. In a month where trial starts doubled, it can swing the figure by fifteen points in either direction.

The fix is to count by cohort: take everyone who started a trial in a given week and ask what became of them. That is how Ascent reports it. The figure for a week is left blank until every trial in that week has had time to end, which means the newest number in your email is always at least one trial length old. People ask us to show it sooner. We would rather show nothing than a number that will be different tomorrow.

It hides what happens next

A trial that converts and is refunded two days later counted as a conversion. So did the subscriber who paid for one month and cancelled. Neither is what you meant by “paid.”

We report three numbers side by side for each cohort:

• Converted: the first payment went through.
• Kept: it was not refunded within 14 days.
• Renewed: a second payment went through.

Across the 1,870 subscription apps on Ascent, the median gap between converted and renewed is 9 points on monthly plans. An app that converts 41% and renews 38% is in good health. One that converts 55% and renews 31% has a paywall that is better at closing than the app is at keeping, and the first number alone would never show it.

It moves when your traffic does

This was the August email. People who find you by searching for exactly what you make convert well. People who arrive from a feature, a promotion or a busy week on social media are more curious than committed. More of the second group lowers the rate and raises the revenue.

So read the rate next to where the downloads came from. If App Store search held steady and the rate fell, look at your paywall. If the rate fell in the same week that browse or referral traffic tripled, you are most likely looking at a good week.

A better morning question

Dana Whitfield, who builds Tidepool, stopped looking at the rate in isolation last spring. Her question now is how many trials reached the paywall at all. Ascent had flagged that a third of Tidepool’s trials ended without the upgrade screen ever being opened, because the reminder that led to it was scheduled for day eight of a seven-day trial. She moved it to day five. Conversions rose 23% with the same paywall, the same price and the same app. Her story has the details.

The rate is still worth watching. It just answers a narrower question than it seems to: of the people who started, in that week, from those sources, how many are still paying. Read it that way and it stops delivering bad news that is not.