Churn Happens Early | Activation || Value Paths

SaaS companies predominantly track churn across their entire user base: how many total users/customers went inactive within a given time period.

In a total-user context, the churn rate is often pretty linear over time: if you lose 2% of your user base in August, you will probably lose another 2% in September, and 2% in October, etc.

When you're focusing on activation, though, churn behaves like a completely different animal.

Activation doesn't care about total users; it only cares about how efficiently you're retaining NEW ones, and new users churn very differently than a full user base that also includes your longest-retained segments.

In other words, the kind of churn that activation focuses on is cohort churn: how long does a new batch of users last before it's gone?

Once you start looking at cohort churn, a big thing that jumps out is that it is NOT flat over time -- there's almost always a steep dropoff at the beginning, then things stabilize later on.

When I see companies undervalue activation, a big part of me thinks that they're assuming that their new users churn linearly, just like their total users do.

To use an example, here's what a cohort's churn would look like if the rate was constant, like air slowly leaking out of a balloon:

If you click on any of the columns, you can see what it looks like if churn was canceled out at that step.

In short: not much of a difference from one step to another.

Every churn-canceled day results in 319 users at the end. Every two churn-canceled days results in 332. The timing of the days themselves doesn't really matter.

In a flat-churn context, investing our efforts toward solving one step's churn works just as well as any other step.

In cohort-land, though, churn doesn't work that way. At all.

Cohort churn is heavily front-loaded, looking more like this:

Whoosh -- big difference, right?

Click on any column to see the contrast of impact if you had solved churn for that step (I bet you can guess the biggest one).

The key takeaway is that instead of one big gentle slope, this has two clear stages: an immediate freefall, and then a leveling-off afterward.

I call this shape the "reverse hockey stick of agony" (not sure what the professional statisticians call it), and it's made up of two parts:

1. Starting churn (where the slope is more pointy-downy than pointy-acrossy) is often where the majority of a cohort's churn is experienced, despite the fact that it happens within a relatively brief window.
2. Sustaining churn (the prolonged, more pointy-acrossy slope) represents the long tail that behaves a lot more like the slowly-leaking balloon that people seem to think all churn is.

In other words (and to push the balloon metaphor to its absolute limit), people seem to think that cohorts churn like "ssssssssssssssss" all the time, but really they churn like "FLPHPFLFPFLFFPFL" at first and only later reach a stabilizing point of "sssssssssssss".

The reason that the "FLPHPFLFPFL" part (aka 'starting churn') is so relevant here is the one-way nature of cohorts: when people churn out, they're gone for good. The air never goes back into the balloon, only out, and a lot of users are gone well before the "sssssssss" (aka 'sustaining churn') part gets to kick in.

In conversion terms, every early loss kneecaps the potential of later-stage cohort levers, and the early losses pile up quickly.

As a simple example, let's say you're trying to improve the conversion rate of your 14-day trial: according to the chart above, even if EVERYONE who made it to Day 14 converted, it would still max out at only 19% because that's all that's left of the signup cohort by that point!

Why does churn concentrate so early? For at least three reasons:

• Low Trust: the earlier the stage of the relationship, the lower the reason to be confident about its future
• Low Cost of Switching: signups are incentivized to fail fast, so they can more quickly find something that works long-term
• Low Intent Support: signups arrive with a different set of goals than power users have, and most product work skews heavily toward the latter

In fact, the chart above is generous: it shows a Day 2 retention rate of 60%, which is on the high end in my experience. Many self-serve companies see fewer than 50% of their signups even log in a second time.

For companies operating along those lines (and perhaps your own), any project that focuses on user actions that take place outside of the first session inherently targets less than half of the original cohort base.

But there's an upside: since churn happens so early, that means that "early" is also where most of the impact often resides.

Putting It into Action

The Day 1-30 example defined its funnel steps in a metronomically time-based way (day by day), but the "churn happens early" observation applies to basically any sequence of user actions (step by step, instead).

For example, let's say that we're Figma and we're trying to juice exposure to a paywall that fires when users hit the "exceeded 3 projects" usage limit.

We pull the numbers, and are presented with dropoff data that looks like this:

Let's say that our pretend-team at pretend-Figma has limited experimentation resources, and has to pick only one stage of the funnel to optimize for unlocking more paywall views.

• It might be tempting to think about ways to get more users who reach Project 3 to attempt a fourth project (and therefore trigger the paywall) -- after all, they're the closest to the end result.

• Or, conversely, it might be reasonable to focus on getting more users who reach Project 2 to create a third, because there are more of them, and they've already demonstrated an interest in creating multiple projects.

• But speaking of "more of them", there are a LOT more users who have only created one project...

• ... and especially more who have never created any projects at all.

If you click on the columns to cancel out each step's churn, the math becomes pretty visibly clear (give it a whirl, it's fun).

Here's what killing all churn for any given step nets us:

Step	Churn% killed	Saved users	More paywall views
Signup → Project 1	60%	600	+164
Project 1 → 2	40%	160	+73
Project 2 → 3	30%	72	+47
Project 3 → View paywall	35%	59	+59

If you ask me, the clear winner here is Signup → Project 1, because it results in 164 additional paywall-viewing users.

For context, that means more than twice as many signups would reach the "view paywall" step as they currently do (compared to current 109).

It would also provide almost as many new paywall-viewers as fixing all the other three stages put together (164 vs 179).

A Big, Fat Caveat

Now, of course, it's complete lunacy to suggest that any funnel step can achieve the exalted nirvana of 0% churn, or to assume that even if you could accomplish it, the additional users would cascade across the rest of the funnel at the exact same percentages that "regular" users currently do.

This especially pertains to the very earliest steps, because many of these percentages are dragged down by users who bounce within seconds of signing up, before they even could do something meaningful (or are bots, etc).

Even acknowledging the above, though, it's still equally unwise to assume that none of the users are salvageable, or that the salvaged ones would necessarily retain any worse across their lifetimes than present users do.

This is especially true if you aren't carefully attending to this vulnerable stage of the user/product relationship to begin with. Which, to be frank, most companies aren't.

In fact, if the additional users were saved because their needs were better addressed than they otherwise would have been, it's actually possible that they will retain BETTER over their lifetimes.

Where the Leverage Lives

This whole argument essentially boils down to two different observation-kids in a trenchcoat:

1. Higher-churn steps provide higher leverage
2. Higher-churn steps tend to show up early

Because churn happens so early, that means "early" is where most of the GOOD things happen, as well.

"Most of the cohort" only sticks around for a relatively brief amount of time, and whether we're talking about value milestones, conversions, referrals, invites -- you name it -- the time range is narrower than most teams seem to treat it.

If you look at all of the activity (good, bad, or otherwise) that a cohort ever produces...

• The highest-performing month will basically always be its first month
• The highest-performing week will basically always be its first week
• The highest-performing day will basically always be its first day

... all the way down to the first hour, and sometimes even the first MINUTE.

Regardless of how deep into the user lifetime your growth lever seems to be, make hay while the sun is shining and your cohort still exists.

For new users in SaaS, that window is briefer than you might think.

To recap...

1. A cohort only shrinks.
2. So the leaving front-loads.
3. So the opportunity front-loads with it.
4. And the early leaving is a cliff, not a slope.

Cohort churn doesn't happen at a slow, steady, inevitable pace.

In reality, we should be scrambling to serve users in every early moment, because most of our signups are gone after "early", and early ends fast.