Win-Back Email Campaign: How to Recover Lapsed Customers in 2026

Customers who quietly stopped buying are usually a larger pool than the ones who never bought at all. They already trusted you once, they know the product, and reaching them costs almost nothing. The catch is that a win-back email campaign is one of the few sends that can damage your entire email programme if you run it carelessly, because by definition you are mailing the oldest and coldest addresses you own.

This guide covers how to define a lapsed customer from your own data, what each email should do, when a discount is the right tool and when it is a trap, and how to measure the outcome honestly.

What a Win-Back Campaign Actually Is

The four terms in this space get used interchangeably, and that is why so many programmes end up mailing the wrong people the wrong message.

A win-back campaign is about purchase behaviour, not inbox behaviour. Someone who opens every newsletter but has not ordered in eight months is a win-back target. Someone who bought last week but stopped opening months ago is a re-engagement target, and a “we miss you” message to them would be plainly wrong.

If your problem is inbox silence rather than a stalled purchase pattern, the re-engagement email guide covers that track, including subscriber-side sequences. If the customer is mid-session with items left behind, that is the abandoned cart email job instead.

Define Lapsed From Your Own Data

The default advice is to treat 90 days without a purchase as lapsed. For most catalogues that number is either far too long or far too short, and it produces two failures at once: customers contacted while they are still perfectly happy on their normal cycle, and customers contacted three cycles after they had already moved to a competitor.

Use your repurchase interval, not the calendar

Take every customer with two or more orders, compute the gap in days between consecutive orders, and look at the distribution rather than the average, which is skewed by a few very fast repeat buyers.

SELECT
  product_category,
  COUNT(*)                                                        AS gaps,
  PERCENTILE_CONT(0.50) WITHIN GROUP (ORDER BY gap_days)          AS median_gap,
  PERCENTILE_CONT(0.75) WITHIN GROUP (ORDER BY gap_days)          AS p75_gap,
  PERCENTILE_CONT(0.90) WITHIN GROUP (ORDER BY gap_days)          AS p90_gap
FROM (
  SELECT
    customer_id,
    product_category,
    order_date - LAG(order_date) OVER (
      PARTITION BY customer_id, product_category ORDER BY order_date
    ) AS gap_days
  FROM orders
) g
WHERE gap_days IS NOT NULL
GROUP BY product_category;

A workable lapse threshold sits around the 75th to 80th percentile of that gap. Past that point the customer is behaving unlike three quarters of your repeat buyers, which is a genuine signal rather than an arbitrary date.

Expect the answer to differ by category

Run the query per category, because a single store often contains several completely different clocks. Ground coffee, contact lenses and pet food have gaps measured in weeks. Skincare and supplements sit in months. Footwear, luggage and kitchen equipment sit in years, and a “we miss you” email three months after a mattress purchase reads as a company that does not know what it sold. For those long-cycle categories the correct win-back is a cross-sell into a consumable or accessory line, or a genuine replacement-cycle reminder, not a plea to rebuy the same durable item.

Handle one-time buyers separately

Single-purchase customers have no interval of their own, so use the distribution of time to second order across your repeat cohort. If most second orders land within 40 days, a customer at day 70 with one order is lapsing. Their message differs too: no habit has formed yet, so the job is proving the second purchase is worth it rather than reminding them of a routine.

Cross the purchase axis with the engagement axis

Purchase recency alone does not tell you how to send. Combine both axes before building segments.

Building those segments means purchase history from the ecommerce platform has to meet sending logic in the email platform, which is an integration problem before it is a copywriting one. Tools such as Tajo keep order data, categories and last-purchase dates synchronised into the sending system so the segment can be expressed there at all.

The Sequence: Three or Four Emails, One Job Each

Time the sequence from the lapse trigger, not from a fixed calendar date, so that each customer enters at the moment that is late for them specifically.

Email 1: the value reminder, sent at the lapse trigger

No discount, no apology, no guilt. Assume the most common reason for silence: the customer forgot, or ran out and bought elsewhere by default. Lead with the specific product they bought and make reordering a single click. Subject line angles here are practical rather than emotional:

• “Time for a refill on the Ethiopian roast?”
• “Your usual size is back in stock”
• “Reorder in one tap, delivered Thursday”

The body should contain their last order, the reorder button, and nothing else. Anyone who returns from this email costs you nothing in margin, which is exactly why the discount cannot come first.

Email 2: address the likely reason for leaving, 5 to 7 days later

This is the email most sequences skip, and it is the one that produces useful information. Name the plausible reasons out loud and make replying easy.

• “Did something go wrong with your last order?”
• “Was the fit not right?”
• “Tell us what to fix and we will fix it”

Send it from a monitored address rather than a no-reply sender, and ask one question only. A customer who replies “the delivery took three weeks” has told you something no dashboard will. Route those replies to support with priority, because a resolved complaint reactivates better than any offer.

Email 3: what has changed, 7 to 10 days later

Give the customer a factual reason to look again: restocked sizes, a reformulated product, a redesigned component that had been criticised, new arrivals in the category they actually bought from, or evidence from customers like them.

• “Six things changed since your last order”
• “The strap everyone complained about has been redesigned”
• “Back in stock in your size, 40 people were waiting”

Personalise the category, not just the first name. New arrivals in womenswear sent to someone who only bought garden tools is noise, and noise from a dormant address generates complaints.

Email 4: the offer or the sunset notice, 10 to 14 days later

Branch here. Customers who are engaged with the sequence but have not bought get a final, clearly time-boxed offer or a preference reset. Contacts who have shown no signal at all get an honest sunset notice instead.

• Offer branch: “One-time restart: 15 percent off until Sunday”
• Preference branch: “Want fewer emails instead of none? Pick what you get”
• Sunset branch: “Last email unless you tell us otherwise”

The sunset variant should contain a single positive-confirmation link, because a click is the cleanest evidence that a human is still behind the address.

The Discount Question, Answered Honestly

A discount is the right tool when price was the plausible reason for leaving, when a competitor undercut you on an identical commodity, or when the margin on the resumed purchase stream comfortably exceeds the one-time cost. It is the wrong tool when it becomes predictable.

The failure mode is measurable. If every customer receives the same 20 percent code at day 60 of silence, then in a replenishment category the rational customer learns to stop ordering at day 55. Compare the next gap length of customers reactivated by a discount against those reactivated without one: if the discount cohort lapses again faster and repeatedly, you have trained the behaviour rather than fixed it.

Alternatives that cost margin only once, or not at all:

• A restock notice for the exact variant they bought
• Genuinely new arrivals inside their purchased category
• A product improvement that answers a known complaint
• Free shipping on the reorder rather than a percentage off the goods
• A sample of a new line added to their next order
• A preference reset that lowers frequency instead of raising the discount
• Expiring loyalty points, a real deadline that costs no price

Vary the incentive between cohorts and time windows so it cannot be predicted, and keep the deepest offer for the last message.

Deliverability: The Part Most Win-Back Articles Skip

A win-back audience is, by construction, your oldest addresses, and abandoned mailboxes do not stay neutral. Spamhaus describes recycled traps as once-valid addresses that providers turned off, where “all mail to these addresses is rejected with a hard bounce for a period of time, often 12 months or more” before being quietly turned back on as spam traps. Hitting them reads as poor list hygiene, not bad luck.

Clean before you send

Remove every address that has already hard bounced, drop role accounts, and validate the oldest slice. Then apply an age cap: if an address has had no opens, clicks, sessions or orders for over a year, the risk usually outweighs the recoverable revenue. Spamhaus notes that trap aging follows an M3AAWG suggestion of 12 months as a minimum, which makes a year a ceiling rather than a starting point.

Stagger the send

The same Spamhaus guidance is blunt about volume: “sending to all unengaged users at once is a sure-fire way to destroy your reputation”, and it recommends staggering the campaign over several days while watching domain and IP reputation. Send in ascending order of risk, starting with the most recently active slice and widening only once the previous batch has settled. A few thousand a day suits a mid-sized list.

Isolate the stream where you can

Send win-back traffic from a dedicated subdomain so any reputation damage stays away from transactional and core marketing mail. A dedicated IP is only worth it if you have the volume to keep it warm, otherwise subdomain separation plus careful batching is the practical option.

Set a stop rule before you press send

Google’s email sender guidelines require bulk senders to keep spam rates reported in Postmaster Tools below 0.3 percent, advise staying below 0.10 percent, and require one-click unsubscribe on marketing mail. Define the pause threshold before launch, watch complaint rate, bounce rate and blocked deliveries per batch, and stop the sequence if any of them climb. A paused win-back campaign costs one week of recovered orders. A damaged sending domain costs every campaign for months.

The Sunset Policy

A sunset policy is the written rule that decides when you stop mailing someone permanently. Spamhaus defines it as “a strategy deployed to rid your email database of old email addresses that are no longer engaging with your email”, and the win-back sequence is the last chance before it applies. A workable rule has four parts:

1. The signals that count as life: opens, clicks, site sessions, app opens, support contacts, and orders.
2. The window: the inactivity period that triggered the sequence, plus the sequence duration itself.
3. The action: move to a suppression list, do not delete, and record the date and reason.
4. The exception: any new order, click or site session automatically removes the suppression.

Recording matters. Suppression should be a stored status with a timestamp, not a deletion, because a deleted contact gets re-imported by the next integration sync and quietly mailed again. Removing these contacts is not a loss. It protects the deliverability of everything else you send, and the cleaner engagement profile benefits every campaign that reaches the people who do respond. The tiered frequency reduction that can precede suppression is covered in the re-engagement email guide.

Measuring What Actually Happened

The metric that matters is not open rate. Opens are inflated by privacy-preserving image prefetching and say nothing about whether a customer came back. Measure these instead:

• Reactivation rate: customers placing an order within a fixed window, for example 60 days from sequence entry, divided by customers entered.
• Retained revenue: revenue from reactivated customers over the following 180 days, which shows whether they resumed a habit or took one discounted order and left.
• Second-order rate among reactivated customers, split by whether they returned with or without an incentive.
• Margin given away, so a campaign that “recovered” revenue at negative contribution is visible.

Use a holdout, not attribution

Some lapsed customers were always going to come back, and counting every returning customer who received the sequence counts them too. The correction is a randomly selected holdout of 5 to 10 percent of eligible lapsed customers who receive nothing, sampled at entry so the two groups are comparable.

Incremental reactivation is the treated rate minus the holdout rate. It is almost always meaningfully lower than the headline number, and it is the only figure that answers whether the campaign caused anything. Run the holdout permanently rather than once, because the answer moves as assortment, pricing and lapse thresholds change. The same test settles the discount question: if the offer branch and a no-offer branch reactivate at similar rates, the discount is margin handed to people who were returning anyway.

Putting It Into Production

Build the campaign as an always-on automation, not a quarterly blast. Entry condition: days since last order exceeds the category threshold. Exit conditions: any order, or entry into the sunset track. Branch on engagement so cold contacts are batched while engaged lapsed customers flow at normal speed. Review the thresholds twice a year, since repurchase intervals shift with assortment and season.

Then connect it to the rest of the lifecycle. A strong welcome email series reduces how many customers lapse at all, drip campaigns carry the ones still deciding, and a broader customer retention programme stops win-back from becoming your busiest automation.

Key takeaways:

• Win-back follows purchase recency, re-engagement follows inbox silence, and the two need different lists and copy.
• Derive the lapse threshold from your own repurchase percentiles per category, not a flat 90-day rule.
• Give each of the three or four emails one job, and delay any incentive until the last.
• Clean, batch and isolate the send, and set the complaint threshold that stops it before launch.
• Judge it on incremental reactivation against a holdout, and on revenue six months later.