How to Build Effective Win-Back Campaigns in Braze

Braze win-back campaigns fail on definition more often than on copy. Someone has to decide what lapsed means, in days and in behaviour, before anyone writes a subject line. Braze gives you two routes: a segment built on last activity, or a churn score that flags people before they go quiet. Pick deliberately, because the two behave differently.

‍

Key Takeaways

• Lapsed is a product question. Braze documents the mechanism for capturing lapsing users and leaves the threshold to you.
• Braze's worked example uses two Last Used App filters so a weekly recurring campaign captures one week of users, not the same people every cycle.
• Predictive Churn flags at-risk users while they are still active, using a Churn Risk Score from 0 to 100.
• A prediction needs a churn definition, a prediction audience, an update schedule and a check of the prediction quality score.
• Intelligent Timing needs engagement data, so set a fallback time for lapsed users.
• Measure return on the action that made the user valuable, not on opens.

‍

What counts as lapsed, and who decides?

That's a product question, and the answer changes by category. A grocery app and a tax filing app have very different silences.

Braze documents the mechanism rather than the threshold. You set up automated recurring re-engagement campaigns to capture lapsing users, and you choose the re-engagement timeframe and recurrence.

Start from your own sessions data. Braze suggests looking at your sessions graph to target users just before high-usage periods. That same graph shows where normal quiet ends and lapse begins.

‍

How do you build the segment without double-messaging?

Here is the detail teams skip. Braze's worked example uses two filters, not one:

‍

The second filter is the important one. Braze explains why: because the campaign recurs weekly, the segment needs to capture at least one week of users. One filter alone would re-include the same person every week until they returned or uninstalled.

Bounding both ends turns an open-ended audience into a cohort. That is what makes a recurring send safe to leave running.

We would extend the same logic with further bounded windows, each starting where the last one ends, so nobody receives two messages in the same week. That is our suggestion, not a Braze example.

‍

When should you use Predictive Churn instead?

The segment approach reacts. Predictive Churn anticipates. Braze's use case describes predicting which users are at risk and sending tailored messages while they are still active.

Setting one up is a series of explicit choices rather than a toggle:

1. Define churn as an event that did not happen inside a window. Braze's example selects "do not" plus a custom event, with a prediction window of 14 days.
2. Choose a prediction audience. The example uses users who triggered relevant events in the past 30 days.
3. Set an update schedule. The example uses weekly.
4. Read the prediction quality score after training, which tells you whether the predictions are likely to be accurate.
5. Build a segment on the Churn Risk Score, which Braze assigns from 0 to 100.

‍

Step five carries a threshold decision. The example targets a score above 70 and uses the prediction audience slider to preview how many users fall into each score range. Move the threshold down and you reach more people, which is a cost question as much as a targeting one.

Both routes can coexist. Prediction catches people who are fading but still present. The last-activity segment catches people who already went quiet, including anyone the model missed.

‍

How do you time and measure Braze win-back campaigns?

Intelligent Timing delivers to each user at the send time Braze calculates as the one they are most likely to engage. The inputs include session times, push direct and influenced opens, email opens and clicks, and SMS clicks where link shortening and advanced tracking are enabled.

There is a catch specific to this audience. Braze lists your most engaged users as a good case, because they have the most engagement data. Lapsed users have the least recent data to offer. Braze lets you specify a fallback time for users without enough data, so set one and expect it to be used more often than on an active campaign.

Measurement needs a behaviour, not an open. In Braze's example, the campaign conversion event is the custom event that represents genuine return, and the Prediction Analytics page compares predicted churners with actual ones.

As a Braze implementation partner, our recommendation is to measure a win-back campaign on the action that made the user valuable. An open rate tells you the subject line worked. It says nothing about whether anyone came back.

The Braze pages we read don't cover how many cycles a lapsed user should receive before suppression. That exit rule is yours to write, and worth writing before launch.

‍

What changed recently?

Two releases touch this workflow. On 28 May 2026, Braze added workspace time zones, so scheduled campaigns and Canvases that don't use local time or Intelligent Timing send according to the workspace's designated time zone. If your win-back campaign is scheduled rather than triggered, check which time zone it now resolves to.

On 25 Jun 2026, Operator became available from the Campaigns page or from within any existing campaign. Braze's example prompts include one about sending lapsed users a push notification. That is useful for a first draft, though the definition question above still belongs to a person.

‍

How does this fit the wider programme?

Win-back is the sharp end of retention, and it works best when the earlier journeys have done their job. An onboarding journey and the activation campaigns around it decide how many users reach the lapsed state in the first place.

Personalisation and dynamic content are what make the message worth opening once the definition and the timing are right.

‍

Frequently Asked Questions
‍

1. How long should someone be inactive before a win-back campaign?

Braze doesn't publish a threshold. Its worked example uses users who last used the app more than 2 weeks ago and less than 3 weeks ago, and it recommends reading your own sessions graph.

2. Why does the example segment use two filters?

Because the campaign recurs weekly, the segment needs to capture at least one week of users. One filter alone would keep re-including the same people.

3. What is a Churn Risk Score?

A score between 0 and 100 that Braze assigns to each eligible user after a churn prediction has trained.

4. Does Intelligent Timing work for lapsed users?

It can, but it relies on engagement data, and lapsed users have less recent data to offer. Set a fallback time.

5. What should a win-back campaign count as a conversion?

The behaviour that represents real return, set as the campaign conversion event, rather than an open or a click.

‍

Sources

• Braze, Capturing lapsing users. Retrieved 01 Oct 2026.
• Braze, Predictive Churn use case. Retrieved 01 Oct 2026.
• Braze, Intelligent Timing. Retrieved 01 Oct 2026.
• Braze, Release notes, 28 May 2026. Retrieved 01 Oct 2026.
• Braze, Release notes, 25 Jun 2026. Retrieved 01 Oct 2026.
• Written by CustomerIK, a Braze implementation partner.

‍

CustomerIK is a Braze implementation partner working across onboarding, technical integration, marketing operations and customer data management.

If your lapsed users keep getting the same message twice, let's talk.

‍