The customers you paid to acquire are leaving. We find out who, act before they go, and prove what it changed.
Retention tools have existed for years. None of them can tell you what they actually caused. We hold a control group, so we can.
Every other retention tool sends a message, watches a number move, and takes the credit. None of them can tell you what would have happened anyway, because none of them hold a control group. We do, on every play, permanently. That one decision is what turns everything below from an opinion into a measurement.
Most of your customers should be left alone. Some stay regardless, some leave regardless, and some leave precisely because you contacted them. Only one group is actually worth reaching, and we can tell them apart.
We describe your paying customers precisely enough that you can stop buying the segment that never converts. For most businesses that saves more than anything we do on the retention side.
You are not billed for retention we cannot show we caused. No difference, no invoice.
The gap between those two lines is not a correlation or a chart that happened to move. It is the difference we made, and it is on your dashboard from the first week.
Shown here from our behavioural model on published industry rates. Your dashboard shows your own two lines, whatever they say.
Every retention tool treats a leaving customer as a leaving customer. But run the same play against a control group for long enough and your customers separate into four groups, and only one of them is worth spending anything on.
Going, not gone. They stay if you reach them, and leave if you do not. The only group worth spending on, and far smaller than the list a churn tool hands you.
They had already decided to stay. Every message you send them is cost without return.
No message recovers them. The list looks urgent and responds to nothing, which is how most retention budgets disappear.
They leave because you contacted them. The message did not fail. It did the damage. Almost nobody knows this group is in their customer base.
A churn score cannot tell these apart, because it only predicts who leaves, not who responds. Separating them needs treated and untreated users measured side by side, which is what our architecture produces by default. Fewer messages, sent to the people they actually work on.
Early on this resolves at segment level, and sharpens toward the individual as your data accumulates. The dashboard tells you which it is rather than implying more precision than it has.
Any business where customers come back, or do not. Subscription products, e-commerce stores with repeat buyers, membership sites, gyms and studios, clinics, courses and training programs, SaaS tools, and consumer apps.
If your customers reach you through a website or a store, Retain installs as a script. If they reach you through a booking system, a point of sale, a mobile app or anything else, that system sends events over the HTTP API. The behavioural engine does not care what the front door looks like.
Measurement, drift detection, plays, widgets, experiments, and checkout on Stripe rails in early access. One integration, both ends of the stack.
See what it does →
Habits take two to five months to form. Every common tactic stops in week one. The evidence, the model, and the taper that follows from it.
Read the evidence →
No charge until your business earns. Compass at $29/mo when you want direction, not just data. Checkout at 1% of what it processes.
See pricing →
Braze, OneSignal, Iterable, real tools, real companies. Every one of them shows you what users did after a message went out. That's a coincidence with a chart on it. We deliberately leave a slice of your customers alone, so what you see is the difference we actually made. It's the one thing the category has never offered, and it's free on every dashboard.
True, useful, and three weeks late. By the time someone cancels, the habit died a month ago. The decision was made long before the billing system noticed.
Winning them back now means re-recruiting a stranger.
Habits die quietly, weeks before the money stops. A missed session against their own pattern is the first symptom, and the last moment intervention is cheap.
Pulling back a drifting user beats resurrecting a dead one 60–70% to 5–20%.
Web apps and PWAs are the cleanest fit. Anything else can send events over the HTTP API. Setup takes minutes and we handle the integration details.