Flattening Retention Curve

For a long time, the first number anyone reached for when asked how the product was doing was signups. Five thousand new accounts this month sounds concrete and copies neatly into a slide. The trouble is that it counts only the people arriving and never the people leaving.

So the same meeting kept happening. Marketing reports five thousand signups, and in the same room finance reports flat revenue. Both are telling the truth and nobody can explain the gap, because nobody is looking at how many of those five thousand were still around a month later. When you keep pouring water in and the level doesn't rise, the thing to examine is the leak, not the volume poured — and a signup count only reports the volume poured.

The retention curve is the chart drawn to see the leak: the share of users still around plotted week by week. Flattening is the follow-up question. A curve falling at all is unremarkable — that happens to every product. What matters is where it stops. If it levels off at some height, that share of people found a place in the product. If it runs all the way down to zero, nothing is holding anyone. Among the ways of judging product-market fit (PMF), this is the one that reads behavior rather than opinions.

Group users by when they arrived, plot what share of each group is still active week by week, and watch for the slope to go flat. Three steps.

1. Cut the groups. Everyone who signed up in the first week of March becomes one block. Counting all users at once lets new arrivals keep washing into the number and blur it, so the work is done as cohort analysis — comparing groups by the period they entered.
2. Compute the share remaining for each week. What counts as remaining is a decision the team makes in advance. Whether a login is enough, or whether someone has to edit at least one document, moves the whole curve up or down. Set the bar loosely and the curve levels off beautifully while meaning nothing.
3. Connect the points and find where the slope approaches zero. Three or four consecutive weeks moving within about a percentage point of each other count as flat at that height.

The numbers make the difference obvious. Start with 100 people and see 45 at week one, 30 at week two, 24 at week four, 23 at week six and 23 at week eight, and the curve has flattened at 23 percent. A run of 45, 30, 18, 9 and 3 is still heading for the floor. Over the first four weeks the first curve doesn't look dramatically better than the second, but a year later the two products are nothing alike.

The height where it flattens matters because that share also applies to everyone you bring in next. A product flat at 23 percent that adds a thousand users a week gains 230 permanent users a week, so the total keeps compounding. A product converging on zero hits a ceiling that no amount of spending moves: the level where arrivals and departures cancel out. Bring in a thousand users a week and lose a thousand a week, and the total sits at whatever number that balance implies, no matter how large the budget behind it gets.

The first place it comes up is the meeting about which stage the company is in. A curve that hasn't flattened means pre-PMF, the stage where demand hasn't been found and the job is still to learn why the few who stayed stayed. Only after it flattens does the post-PMF question — how to acquire faster and cheaper — make sense. Reverse the order, and work spent lowering customer acquisition cost (CAC) before the curve flattens is work spent finding cheaper water for a leaking bucket.

The second is segmentation. It's common for the overall curve to slide to the floor while one slice levels off. At Pagedeck, week-four retention across all accounts was 11 percent, but among teams where five or more people edited the same document, the curve flattened at 38 percent. That slice becomes the beachhead market — the first market to win before expanding sideways.

The third is deciding whether an experiment worked. Judging an onboarding change or a new feature by first-week numbers usually misleads. Sending more notifications lifts week-one retention while the height where the curve settles stays exactly where it was. The part to read is not the front of the curve but the place it lies down.

"It's been flat at 24 percent since week four. We can turn acquisition up now."

The share who stick has been confirmed, so money put into marketing won't drain straight back out.

"The curve hasn't flattened yet, so hiring feels early."

A brake: demand isn't confirmed, so don't add fixed costs.

"It doesn't flatten overall, but it does if you look only at team accounts."

A finding that the product fits a particular kind of buyer, and a proposal to revisit who the company sells to.

Mia runs a meeting around this curve once a quarter at Pagedeck. Last quarter, when Ravi proposed doubling the ad budget, what she brought was the curve split by cohort. The individual-signup curve was still falling at 4 percent in week eight with no sign of stopping, while accounts that arrived as teams had flattened at 31 percent.

So the budget went up and the target narrowed to team accounts. The same money spent on individual signups leaves almost nothing eight weeks later, and the chart said so plainly.

The most common misreading is treating any flattening as a pass. The word sounds binary, as though the curve either clears the bar or doesn't. In practice the height matters as much as the shape. A curve flat at 3 percent has flattened, and that product runs while losing 97 percent of everyone it acquires. Ignore the height and you learn the arithmetic doesn't work only after the ad money is gone.

The other is reading a flat curve as nobody leaving. The line looks horizontal, so churn seems to have stopped, but some of the remaining users still drift away every month — the count is just small enough to disappear into the graph. Treat that plateau as permanent and let existing customers go unattended, and over a year or two the line that looked flat sags quietly downward.