Churn Is a Lagging Indicator: What to Watch Instead

A renewal is three weeks out and the account looks healthy. Invoices paid on time, tickets closed, nothing hostile in the inbox. Then the customer does not renew, and the post-mortem lands on a reason everybody nods at and nobody can act on: budget.

The decision was not made three weeks out. It was made months earlier, on a day nobody logged, when one person stopped opening the product and started working around it. Churn is a lagging indicator: it records an outcome long after the conditions that produced it stopped being reversible. Measuring it more precisely does not make it arrive any earlier.

This piece is about the churn leading indicators that do move early: usage decay, support sentiment, seat drift, and the champion leaving. Each has a shape you can recognise and an intervention that fits it. Each is also available to a small team with a database and a recurring hour, which matters, because most of the tooling sold for this problem quietly assumes a customer success department you do not have.

The position this defends: a save at the renewal call is usually theatre. By then you are negotiating price against a decision that was made on product experience, and price was never the thing that changed. Retention work that starts at renewal is not retention work. It is procurement with a friendlier tone.

What churn actually measures

Churn is an accounting event. It fires when a contract ends or a card stops being charged, and the date attached to it is the date of the paperwork, not the date of the decision. Everything useful sits upstream of that date, in a period where nobody was watching because nothing had gone wrong yet.

Two things follow. A churn rate is a scoreboard rather than a diagnostic: it tells you the season is going badly, and it cannot tell you which week turned, because by the time it moves several unrelated failures have been averaged into one number. And the interventions available at each stage get weaker as the stage gets later.

Watch how the ladder works. Something is wrong with how a team uses the product: fixable in a call. Something is wrong and they have built a workaround: harder, because the workaround now has an owner and a habit around it. They have stopped mentioning the product internally: nearly finished, because nobody is defending the line item. The renewal notice arrives: you are discounting. Same account, four very different conversations, and only the first one is cheap.

Usage decay: the signal that moves first

Usage is the earliest thing to move, and the way it moves is the point. Accounts rarely fall off a cliff. They thin out.

Read the shape, not the total

Total sessions is a poor measure because it hides substitution. An account can log a steady number of sessions while the actual work migrates to a spreadsheet, because the sessions that survive are the ones nobody bothered to replace: an export, a login to check one number, a monthly report that a finance process happens to depend on. The graph looks flat. The relationship is already over.

Four things to watch instead of the total.

• Breadth of features touched. An account using six parts of the product is far harder to replace than an account using one, even when the one gets used daily. Narrowing breadth is the clearest early slope there is.
• Depth per active user. Ten people doing one thing each is a pilot that never ended. Three people doing everything is a real workflow, and also a concentration risk with three names on it.
• The gap between sessions. Weekly becoming fortnightly becoming monthly is a slope worth a phone call. A single quiet week is weather, and treating it as a signal trains everyone to ignore the ones that matter.
• Whether the core action still happens. Every product has one action that means value was delivered. Track that one alone, and never let it be averaged into a general activity number where it can be propped up by logins.

The intervention that fits

Call the account while the slope is still shallow, and do not call it about the renewal. Ask what changed. The answer is usually specific and often small: a report they cannot build, a permission that blocks the person who does the actual work, an export somebody now does by hand every Thursday. Those are product improvement tickets rather than commercial ones, and a customer will hand them to you for free as long as the conversation is not about money.

Decay also exposes a failure that has been sitting there since week one. If usage never reached a plateau before declining, but instead started low and drifted lower, this is not decay at all. The account never landed. That is a different problem with a different fix, and it belongs to what happens in the first ten minutes, not to month seven. Teams that do not separate the two spend retention budget on accounts that were never onboarded.

Support sentiment, not support volume

Ticket volume is the most misread number in the business. A rising count can mean the product got worse, or it can mean adoption reached a less patient department. A falling count reads as good news and often is not, because customers who have given up stop filing tickets. Silence is cheaper than complaining when you have already decided to leave.

Read the tone and the mix instead. Three patterns matter.

• The question that stopped being asked. A customer who used to ask how to do things better, and now only reports things that are broken, has stopped investing in the tool.
• Escalation without heat. Polite, procedural, carefully worded, with a manager copied in. That is somebody building a record, not somebody asking for help.
• Requests that describe a different product. When asks only make sense inside a rival’s model rather than as an extension of yours, the account has already been shopping.

That third pattern needs a caution attached. A feature request is not automatically a product gap, and building the thing will not always fix the reason it was asked for. Sometimes the request is a symptom of a positioning problem wearing a feature request as a costume. Shipping to silence a request is one of the more expensive mistakes on the menu, because you pay for it every month afterwards in maintenance.

Support content is an intervention in its own right and a badly underrated one. A user who cannot find the answer at eleven at night files a ticket if they still care, and opens a competitor’s site if they do not. Treating documentation as a growth channel also treats it as a retention channel: the same page keeps an existing user unstuck and brings a new one in.

The constraint is publishing speed. If a documentation fix takes a sprint, nobody writes one in response to a ticket, and the knowledge stays buried in the reply. A workspace like Acrosite takes the other approach: an editor writes the page in a structured interface, the workspace generates the files, commits them to GitHub and triggers the configured deployment, so the answer that unblocked one customer is public before the next one asks the same thing.

Seat drift and the quiet contraction

Seat count is the most honest number in an account and the least watched between renewals. It moves in small steps, and every step has a reason that somebody inside the customer could explain in one sentence if asked.

Three shapes are worth wiring an alert for. Seats bought and never assigned: a purchase made on optimism that the rollout never justified. Seats assigned but never activated: the product reached a department that did not want it, usually because somebody else chose it for them. And seats quietly not re-added when a person leaves the customer’s business: the slowest of the three, and the most predictive, because it means the product has stopped being part of how a role is defined.

None of this shows up in revenue until the renewal, which is exactly why it is useful. A contract that is going to shrink shows the shrinking for months, provided somebody looks at assignment data rather than billing data. Most companies put billing on the dashboard and leave assignment in the admin panel, then express surprise at a downgrade.

The intervention is not a discount and it is definitely not a webinar. Find the person who stopped being replaced and ask what their role does now instead. Sometimes the workflow moved to another team, and you have a new buyer to go and meet. Sometimes it moved into a spreadsheet, and you have a product problem with a date on it. The routine half of that conversation can be carried by lifecycle programmes that fire on non-activation rather than on a calendar. The other half needs a person on a call.

The champion leaving is the loudest signal nobody hears

One person usually decided to buy. They defended the line item, ran the internal rollout, wrote the short guide their colleagues actually read, and explained in meetings what the product was for. When that person leaves, you do not simply lose a contact. You lose the only account of why the product exists that anybody inside the customer ever believed.

So treat a champion’s departure as the highest-priority event in your retention calendar. Higher than a poor survey response, higher than a spike in tickets. It is the one signal that starts a countdown on a date you can actually know.

It is also detectable without anything sophisticated. An admin email starts bouncing. A title changes on a public profile. Somebody new appears in a support thread with no context and asks a question the account answered eighteen months ago. Each of those is a trigger you can wire in an afternoon, and none of them needs a data platform or a vendor.

What to do about it: re-onboard from the beginning. Not a check-in call. An actual walkthrough of what this product does for this company, with the successor, in their language, handing over the reports and configurations their predecessor built. Then meet whoever now owns the budget line, and rebuild the internal case before they build their own case for cutting it. The successor will otherwise inherit a tool they did not choose and cannot explain, which is a poor position from which to defend a renewal.

Here is the honest limitation. Sometimes there is nothing to save. A champion leaves, a successor arrives carrying a mandate to consolidate vendors, and no amount of re-onboarding beats a decision that was made about the number of suppliers rather than about you. You will lose some of these regardless. The reason to run the play anyway is that the two cases are indistinguishable from outside, and finding out costs one hour.

The churn leading indicators, side by side

Here is the argument as a table. The right-hand column is the part most churn dashboards leave out, and it is the only column that changes an outcome.

The causes nobody puts on a dashboard

Two churn causes almost never get named in a post-mortem, because neither produces a complaint. They produce a shrug, and then a habit.

Interaction latency

A product that responds slowly does not generate a ticket. It generates avoidance. The user learns which screens are painful, routes around them, and their workflow narrows until the product is doing one small job badly. Months later that reads as usage decay, and the post-mortem calls it poor adoption. Interaction to Next Paint is the measurement that catches this, and it belongs on the same review as the retention numbers rather than in a separate engineering report nobody opens.

The fix is unglamorous: identify the three interactions that carry the core action and make those fast before adding anything new. While you are in there, read the words on the screen, because a slow screen and a confusing one produce the same avoidance behaviour, and microcopy is the far cheaper of the two fixes.

Barriers you will never be told about

The second one is accessibility. If part of the product cannot be operated by keyboard, or a status message is never announced, the people affected do not file a bug. They ask a colleague to do that step for them, then stop being users, and the seat gets reclaimed at the next renewal without anyone connecting the two events. WCAG 2.2 is the reference, and the relevant point for this argument is that failures of this kind are silent by construction. Nobody reports a barrier they have already learned to route around.

Why the save at renewal usually fails

Now the uncomfortable part. Most renewal saves get recorded as saves and are not.

A discount at renewal buys a term. It does not change the fact that the product stopped being used, so it converts one churn event into another churn event a year later, at a lower price, with a customer who has now learned that the list price was negotiable. If usage did not recover, nothing was saved. Measure saves by whether the core action came back, not by whether the invoice went out.

A renewal you had to win with a discount is a churn event with a delay attached and a worse margin.

There is a version of the renewal conversation that does work, and it is the one where the commercial discussion sits downstream of a fix that already shipped. "We changed the thing you told us about in March" is a completely different sentence from "what would it take to keep you". The first is retention. The second is a negotiation you entered without a position.

This is where small teams get caught, because the fix that would hold an account competes directly with the roadmap that wins new ones, and both want the same week. That trade-off is real and has no clean answer. It is the same tension as running client work alongside product work: the urgent thing and the compounding thing are never the same thing, and pretending otherwise moves the cost around instead of removing it.

Where we would start on Monday

None of this needs a customer success platform. It needs four queries, three triggers and a recurring hour that does not get cancelled when something louder happens.

1. Define the core action. One event that means value was delivered. Write it down, get two people to agree, and stop there. This is the hardest step and everything else depends on it being honest rather than flattering.
2. Chart it per account, weekly, as a slope. Not a total, and not a rolling average across the customer base, where one account’s decline hides comfortably inside another account’s growth.
3. Wire three triggers. Core action absent for two consecutive periods, a seat unassigned or unactivated past its first week, and an admin contact that bounces. Each one creates a task for a named person, never an automated email.
4. Read tickets for tone, not count. One person, once a week, looking only for the shift from questions towards reports. It takes a quarter of an hour and outperforms most sentiment tooling.
5. Run one diagnostic call a week. Not a business review. Twenty minutes with an account showing a slope, about their workflow, with the commercial relationship deliberately left outside the room.

Instrumenting that properly is analytics work with a small a: event definitions, an account dimension, and one place to look at them. It is well inside the reach of a team of five, and it beats a purchased health score for a simple reason. You will act on a number you defined and argue with a number you bought.

One condition reverses everything above. If your product is bought once, used intensely for a defined period and then legitimately finished, usage decay is not a warning at all. It is the customer succeeding. Reading a slope as risk in that case produces a stream of calls that irritate perfectly happy people. So know which shape your product has before wiring any triggers, and if that is genuinely unclear, the uncertainty is itself the finding: our parent company has written about whether you are selling a product, a service or a feature, and the answer changes what a healthy usage curve is even supposed to look like.

If you want a second read on which of your signals is genuinely predictive rather than merely available, tell us what you are seeing and what your renewal conversations currently sound like.