Build a win-back journey that proves its lift (with a holdout)

A win-back journey re-engages customers who've gone quiet. It's one of the highest-ROI automations you can build — but only if you can prove the revenue it "recovered" wouldn't have come back on its own. That proof is a holdout, and it's the step most teams skip. Here's the whole thing.

Step 1 — Define "lapsed" with a computed trait

Don't guess. Create a computed trait like days_since_last_order and define lapsed as, say, 60–120 days quiet (past 120, a different message). A trait updates as behaviour changes, so the segment maintains itself. In WeZend you can define it in a sentence via the AI copilot, then segment on it.

Step 2 — Reserve a holdout

Before the journey sends anything, randomly withhold 5–10% of the eligible audience. They receive nothing. This is your control group. At the end you compare purchase rate and revenue-per-customer between the treated group and the holdout — the difference is the incremental lift, the only number that actually belongs to the campaign. WeZend builds the holdout into the journey; you don't manage a spreadsheet.

Step 3 — Build the branches

A solid win-back has two or three steps, not one:

1. Soft nudge (day 0). No discount yet — remind them what they liked. "We saved your favourites." Email is fine here.
2. Wait for purchase (up to 96h). If they buy, they exit the journey. Don't keep messaging a customer who already converted.
3. Incentive (day 4). Now a real reason: a time-boxed offer. Consider SMS or RCS for immediacy on high-value segments.
4. Last call (day 7). Short, urgent, then stop. Chasing further trains people to ignore you.

Step 4 — Respect the guardrails

Frequency caps, quiet hours and suppression apply to every step. A win-back that violates someone's opt-out or blasts at 2 a.m. costs you more than it recovers.

Step 5 — Read the holdout, then decide

After a full cycle, look at treated vs. holdout. If the treated group's revenue-per-customer beats the holdout by more than the campaign cost, scale it. If it doesn't, the "recovered" revenue was mostly customers who'd have returned anyway — fix the offer or the timing before you spend more. This is the discipline that separates automations that look busy from automations that pay.

You can draft this whole journey in plain language via Connect AI — "build a two-step win-back for customers lapsed 60+ days, with a 10% holdout" — review the draft, then activate it yourself.

Start free and build your first one this week.