Judge freemium conversion by how long users have stayed, not by month one

Freemium products often look weak in month one because few new users pay right away. This tactic reads conversion by signup cohort as each cohort ages. Evernote reported that only 0.5% of a monthly signup cohort went premium in its first month, and that share climbed to 5.5% after two years. Its CEO said long-time users are more likely to convert.

How it works

The idea is simple. A user who stays for a year has more reason to pay than a user who signed up last week. If you judge freemium on month-one conversion alone, you undercount the revenue that older users will bring. The steps:

• Keep the core free. Keep the main product free and add paid features slowly. Evernote described this approach, with the goal of reaching positive gross margin.
• Set a modest target. Evernote aimed for a single-digit percentage of users to pay a few dollars a month.
• Group by signup month. Put users into a cohort by signup month. For each cohort, record the share of the original signups who have ever gone paid, at the same fixed ages every time.
• Plot the curves side by side. In Evernote's data, a cohort's premium share went from 0.5% in its first month to 1% after six months, 2% after a year and 5.5% after two years.
• Read the curves next to cost. Evernote's CEO wrote that gross margin rose each month because revenue per user grew while variable cost per user fell.

The Evernote case shows the result at company scale. In a 2010 interview, Evernote's CEO gave end of January 2010 figures: 2,335,676 total users and 41,598 premium users. That month, variable expenses were $68,641 against roughly $145,000 in revenue from active premium users. He put gross margin at about 53%. Those variable costs covered about 60 servers and four operations staff.

What it takes

Our estimate: about 16 hours of work, and about 90 days before you can read a signal. Most of the hours go into a cohort query that joins signup date to first payment date, plus a simple chart and a cost-per-user line. If you already store both dates, the query is quick, and the rest of the time goes to cleaning out test accounts and refunds. The signal is slow because the whole point is to watch cohorts age, so you need at least a few months of new cohorts before you can compare shapes. If you have older history, backfill it and you can read the first curves the same week.

What to measure

Cumulative paid share by cohort age is the core metric. A good reading is a curve that keeps rising after month one. A bad reading is a curve that goes flat after the first few weeks, which means waiting will not help.

Compare month-one and month-twelve paid share for the same cohort. If the gap is large, month-one conversion is a poor proxy for what freemium will earn, and you should stop judging pricing changes on it. If the gap is small, your conversion happens early, and a free trial may fit better.

Read free user retention next to the paid share curve. Late conversion only exists if users stay. A retention curve that flattens at a healthy level supports patience. One that keeps sliding toward zero does not.

Track variable cost per user and gross margin. Waiting on old cohorts is only affordable if serving free users is cheap and getting cheaper. For outside reference points on conversion rates, see the free-to-paid conversion benchmarks, but compare yourself to your own past cohorts first.

When it backfires

• Short retention. Libin's own conference talk said cloud freemium works when long-term retention is strong. If users leave within weeks, no cohort ages long enough to convert.
• A product that does not gain value over time. The same talk named a product that grows in value as a condition. A tool used once has nothing that makes month twelve more compelling than month one.
• High variable costs. Low variable costs were the third condition. Evernote's gross margin only turned positive in January 2009, after a closed beta in February 2008. A company with expensive free users may run out of cash before old cohorts pay off.
• Using patience as an excuse. A rising curve can hide a broken first week. Keep working on activation even while you wait for cohorts to mature.
• Mixed cohorts. If you change pricing or the free tier midway, later cohorts are not comparable to earlier ones. Mark the change on the chart.

Run it as an experiment

We recommend this hypothesis: "Our paid share at month six is meaningfully higher than at month one, so month-one conversion understates the value of freemium." Use a before and after comparison with a fixed window. Pick the cohorts from a set span of signup months and compare each at month one and month six.

To test an action, not just measure, run a holdout. Show upgrade prompts to users past a tenure threshold you choose, and hold back a random share of eligible users. Use the sample size calculator before you start, because conversion rates this low need large groups.

Log signup date, first payment date, plan, last active date, refunds and any pricing changes. Our suggested decision rule: if the month-six paid share clearly exceeds month one across several cohorts and retention holds, judge freemium on aged cohorts from now on. If the curve is flat, revisit the free tier. Write it up with the template in how to run a growth experiment, and model the cohorts in the growth model template.

How strong the evidence is

The grade is case with numbers: one company documented the tactic and published results as numbers, through a named executive. That case is Evernote, where the CEO shared cohort conversion figures, user counts and margin. A second archived copy of the interview repeats the same strategy and figures.

The evidence has limits. The numbers are self-reported by the CEO, from one company, covering 2008 to 2010. There is no control group, so we cannot separate tenure from the paid features Evernote added along the way. The cohort figures are an average shape, without per-cohort sizes. The margin gain is also tied to falling server costs, not only to conversion.