How should we segment lapsed users and sequence a reactivation program that doesn't just burn discounts?

Start by redefining "lapsed"

A flat 90-day cutoff lumps very different people together. Someone who ordered weekly and stopped is a much stronger signal than someone who ordered twice last year and is simply on their normal rhythm.

• Measure lapse relative to each customer's own cadence. For example, flag someone when their days since last order exceed 2–3x their median gap between orders.
• Add a separate pre-lapse stage, such as 1.5x the normal gap, and intervene there. Winning back a customer at 30–45 days is far cheaper than at 120.
• Benchmark your 40% against category norms. A meaningful share of one-time and occasional users churn in any delivery app. The goal is to recover the recoverable ones profitably, not to drive that number to zero.

Segment on three axes

Value tells you how much to spend. Reason tells you what to say. Reachability tells you where to say it.

Historical value:

• High: frequent orderers, high AOV, long tenure, low refund rate. These justify human-feeling outreach and, if needed, real incentives.
• Mid: regular but modest. Use automated sequences with conditional offers.
• Low: one or two orders, often acquired with a promo. Spend almost nothing here. Use cheap channels and no escalating discounts.

Inferred lapse reason. Your data can usually suggest this:

• Bad last experience: late delivery, missing items, refund, cancellation, or a low rating on the final order. These customers need an apology and proof you've fixed things, not a coupon.
• Promo-dependent: most orders used a discount, and they stopped when promos stopped. They're likely to take a discount and leave again, so treat them carefully.
• Supply loss: a favorite restaurant left the platform, changed hours, or raised prices. Send them a curated alternative.
• Moved or out of zone: recent address changes, or app opens from outside your cities. Suppress them, or flag them for when you launch in their new city.
• Competitor or habit drift: no negative signal, and order frequency simply decayed. Respond to these with novelty and convenience.
• Seasonal or life change: patterns like students leaving in summer, or ordering that stops at the same time each year. Time outreach to their historical return window.

Current engagement:

• Still opening the app or browsing without ordering. These are warm, and something specific is blocking them, often price, fees, or selection. Use in-app messages and push.
• Opening emails but not the app. Email is your channel here.
• Fully dark: app uninstalled and emails ignored. Reach them through paid retargeting audiences or SMS if you have consent, or accept the loss.

You don't need every combination. Five to eight practical cells usually capture most of the value.

Make discounts the last step, not the first

Lead with non-price reasons to come back, and escalate only for customers worth it who didn't respond.

• Relevance first: "Three new Thai places opened near you," "Your usual spot now delivers until midnight," or "Here's what people in your neighborhood ordered this week." Personalize by cuisine and favorite restaurants.
• Fix friction: if fees were the issue, a free-delivery offer usually costs less than 20% off and lands better. If the issue was a bad experience, acknowledge it specifically.
• Use cheaper offer structures when you do discount. A fixed amount off with a minimum basket beats percentage off. Free delivery or a membership trial builds habit. Restaurant-funded promos let partners carry the cost, and many will for customer acquisition.
• Offer expiry and single-use codes prevent leakage to coupon sites.
• Cap total incentive spend per customer per year, and exclude known promo abusers.

A sample sequence for a mid-value, no-negative-signal lapser

Day 0, triggered at about 1.5x their usual gap:

• A push or email showing what's new near them, sent at the day and time they habitually ordered. No offer.

Day 5–7:

• Social proof or curated content, such as top-rated dishes this week in their favorite cuisine. Still no offer.

Day 14:

• A modest incentive, such as free delivery on their next order, valid for 7 days.

Day 28:

• Only for those whose predicted value justifies it, a stronger offer: a fixed amount off with a minimum basket, or a membership trial.

Day 45+:

• Drop to a low-frequency nurture cadence of monthly content. Stop paying to reach them.

After a successful reactivation:

• The first order back is the fragile moment. Make sure it goes well by prioritizing dispatch where you can.
• Follow up with a second-order nudge within their normal cadence. Reactivation that doesn't produce a second order is mostly wasted spend.

Variations by segment

• Bad experience: send a specific, human-sounding apology within days of the lapse signal. Say what went wrong, offer a credit if it wasn't already refunded, and mention operational improvements. High-value customers here might warrant a personal note from support.
• Supply loss: "We know you loved X. Here are three similar places customers like you switched to."
• Promo-dependent, low-value: one free-delivery touch at most, then nurture only. Don't train them to wait for 20% off.
• High-value, any reason: compress the timeline. Combine the relevance and service-recovery messages early, and allow a richer offer sooner, because each week of delay lowers the odds of winning them back.

Measure incrementality, not response

Your current email likely looks bad partly because there's nothing to compare it to. Some of those customers would have returned anyway.

• Keep a random holdout of 10–20% in every segment that receives nothing. The metric that matters is reactivation in the treated group minus reactivation in the holdout.
• Track 60–90 day retention and contribution margin after reactivation, not just the first order.
• Calculate cost per incremental retained customer by segment. That number tells you where to spend more and where to stop.
• Test one variable at a time per segment: message angle, offer type, timing, channel.
• Once you have a few cycles of data, you can build a simple uplift model. Then you send offers only to people whose behavior they change, not to sure things or lost causes.

Fix the leaks upstream

Win-back data is a diagnostic. If lapse clusters around late deliveries in one city, specific restaurants, fee increases, or promo-acquired cohorts, those are operational and acquisition problems. Feed them back to ops, partnerships, and the growth team. Reducing new lapses is cheaper than any reactivation program.

What to build first

• Cadence-based lapse definitions and a pre-lapse trigger.
• Tagging for value tier and likely lapse reason.
• Holdout groups.
• Two or three distinct sequences: service recovery, high-value, and a general sequence with escalating offers.

That alone should outperform the single 20% email, and it will show you where to invest next.