Welcome to ChurnJitsu, your weekly briefing on the art of retention.

Let’s dive in.

You paid for every install. Here is how to keep more of them.

Your retention chart shows who already left. By the time the dip appears, the user is gone, the acquisition spend is gone, and nothing you ship this week can win them back.

The user worth your attention is still technically active but quietly packing a suitcase. They opened the app three times last week and once this week. They stopped using the core feature. Their last session ended on an error. Your dashboard counts them as retained, and they are already halfway gone.

The scale of the problem is big. AppsFlyer's 2025 uninstall report puts the global Android uninstall rate at 46.1% within 30 days for 2024, with the United States at 48.02%. It also found that the largest wave of uninstalls lands on Day 1.

Waiting until Day 30 to diagnose churn is like checking whether the patient still has a pulse after the ambulance leaves.

What you get from an Uninstall Risk Score

A simple score gives you three things. You see at-risk users days before they vanish. You stop spending pushes, discounts and engineering time on the wrong people. And you find out which broken moment is costing you the most users, so you fix that first.

Skip the 47-feature machine learning model. Start with four signals you probably already have.

1. Activation: did they reach the payoff?

Did the user hit the core value event? For a game, that might be the first meaningful level. For a finance app, a linked account. For a photo app, something generated and saved. A user who never gets there starts behind, and they are the easiest ones to lose.

2. Recency: when did they last get value?

Count meaningful sessions only, like finishing the core action. A user who completed it yesterday is in a different place than one who opened the app from a push notification and landed on a dead screen.

3. Frequency: is their usage holding or shrinking?

Compare each user to their own earlier behavior. Three sessions a week can be healthy in one app and a five-alarm fire in another, and a company-wide average will hide the difference.

4. Friction: did something ugly happen?

A crash, a failed payment, a permission denial, a slow load, a search with no result, or a support ticket with the emotional tone of a hostage negotiation. Users rarely complain twice. They uninstall.

You can pull most of this from tools you already run. Google Analytics for Firebase lets mobile teams define custom events and analyze user behavior, and Firebase surfaces crash, notification, deep-link and purchase data across its analytics stack.

Version one, built in an afternoon

Give every user a score and update it as they use the app:

Pick your own cutoffs for low, medium and high after a week of watching real users.

Turn the score into saved users

The score does nothing on its own. The response is what keeps people.

Low risk: leave them alone. Nobody wants a nudge while things are going well.

Medium risk: surface useful content, visible progress or a clear reason to return.

High risk: change the experience now. Shorten the path to value, recover failed progress, offer help. And stop sending a generic "we miss you" push to someone whose last session crashed three times.

That last step is where most teams lose the users they could have kept.

Keep it honest over time

Compare the score against actual uninstall and retention outcomes. Reweight it, drop the signals that predict nothing, and add the behaviors that matter in your app. Eventually this can grow into a proper model. A crude score that changes the product today beats a perfect churn model sitting in a quarterly analytics deck.

Your move this week: list the five signals above, confirm you can pull each one per user, and write down what a high-risk user should see instead of your current default. That list is your version one.

Don't just measure who left. Build the app to notice who is leaving.

Inman