A DTC client of mine had 40,000 customers who hadn't bought in over a year, sitting untouched in her list because "win-back" had turned into a single generic 20%-off blast twice a year. That's not a strategy. That's a coupon with a mailing list attached.
Before writing any win-back copy, use AI to segment your lapsed list by real signals, time since last purchase, historical spend, and what they bought, rather than treating every quiet customer the same. Lapsed customers are 5 to 7x cheaper to reactivate than acquiring a new customer, and customers who do come back tend to spend more over the following year than first-time buyers. Write win-back copy that references what someone actually bought, not a generic “we miss you,” and always give an easy, honest way to opt out entirely rather than escalating the discount every time they don’t respond.
A DTC client of mine had roughly 40,000 customers who hadn’t purchased in over a year, sitting completely untouched in her email list, because her entire “win-back strategy” was a single generic 20%-off blast sent to everyone twice a year, regardless of who they were or what they’d bought. That’s not a win-back program. That’s a coupon with a mailing list attached to it.
This is the norm, not the exception. Most brands treat every lapsed customer identically: same subject line, same discount, same timing, whether they bought once for $30 or were a repeat $400-a-quarter customer before they went quiet. That flattens exactly the signal that would tell you who’s actually worth re-engaging, and with what.
AI’s real contribution to win-back isn’t a wittier subject line. It’s doing the segmentation and personalization work that nobody has time to do manually across a list of tens of thousands of lapsed customers, so the message and offer actually match who someone was as a customer before they disappeared.
The opportunity here is bigger than most marketing teams treat it. Research on first-purchase behavior suggests a large share of first-time buyers, commonly cited around 70 to 80%, never make a second purchase, which means most brands are sitting on a large pool of “lapsed” customers who were never really activated in the first place, not just customers who drifted away after being loyal.
On the reactivation side, automated win-back sequences typically bring back 12 to 18% of lapsed customers, and Klaviyo’s own published benchmark data, based on roughly 183,000 brand accounts, puts average win-back flow revenue at $0.84 per recipient[1]. That’s not a huge number per email, but multiplied across a list of 40,000 dormant customers, it’s real, previously unclaimed revenue sitting in a segment most brands never systematically work.
The cost math is the part worth internalizing: reactivating a lapsed customer is widely estimated at 5 to 7 times cheaper than acquiring a brand-new one, and customers who do come back tend to purchase more over the following year than a typical first-time buyer. Win-back isn’t a nice-to-have side campaign. For most ecommerce brands, it’s underpriced revenue sitting in a list nobody’s segmenting properly.
Before you write a single win-back email, pull the actual number: how many customers in your list haven’t purchased in 6, 12, and 18+ months, and what did they spend when they were active. That number alone usually reframes win-back from a nice-to-have into a priority.
This is the highest-leverage use of AI in the whole workflow, and it’s not the copywriting. Export your lapsed customer list (anyone who hasn’t purchased in your typical repurchase window, commonly 60, 90, or 180 days depending on your category) with purchase history, and ask AI to help you group them into segments based on real signals, not just “time since last order.”
Something like: ‘Here is a CSV of customers who haven’t purchased in over 90 days, with their order count, total spend, last product purchased, and days since last order. Group them into three to four segments based on likely reactivation approach: high-value lapsed (worth a stronger offer), one-time-only buyers (never really activated, need a different message than a discount), and recently-lapsed vs. long-dormant (different urgency and tone). Explain your reasoning for each segment.’
The output won’t be perfect, and you shouldn’t treat it as final. Its real value is turning an unmanageable spreadsheet of 40,000 rows into three or four groups you can actually write distinct campaigns for, instead of either doing nothing or sending the same blast to everyone.
Once you have real segments, AI is genuinely useful for drafting copy fast, as long as you feed it specifics instead of asking for generic “we miss you” language, which is exactly the kind of message people have learned to ignore.
A prompt that produces usable drafts: ‘Write a win-back email for [segment description] who last purchased [product] on [date]. Reference that specific purchase naturally, don’t open with a discount, and end with either a specific reason to return or a clear, no-guilt way to unsubscribe. Keep it under 120 words.’ Always read the output back and cut anything that sounds like it’s begging. If you wouldn’t send it to a friend, don’t send it to 10,000 people either.
Back to the client with 40,000 dormant customers and a single generic 20%-off blast twice a year. The AI segmentation pass split that list into three groups: about 6,000 high-value lapsed customers (three or more past orders, average spend over $150), roughly 22,000 one-time-only buyers who’d never come back after their first order, and the rest scattered across long-dormant accounts over 18 months old.
The high-value segment got a no-discount check-in referencing their specific product category, restocked items, and a loyalty-tier note about what perk they’d get at their next order. The one-time-only segment, which was never really “won back” since they were never really activated, got a completely different message built around the brand’s actual differentiators rather than a discount, closer to a second first impression than a reactivation pitch. The long-dormant segment got the honest “it’s been a while” email with a clear opt-out.
Six weeks later, the high-value segment reactivated at just over 21%, meaningfully above the 12-18% general benchmark, which tracks with the idea that better-targeted messaging to genuinely valuable customers outperforms a generic blast. The one-time-only segment reactivated closer to 9%, lower but still real revenue from a group that had been getting the exact same ignored 20%-off email for two years running. The total campaign, across all three segments, generated more revenue in six weeks than the previous year’s two generic blasts combined, without a single discount larger than what the brand had already been offering blindly to everyone.
Timing matters more in win-back than in almost any other email type, because you’re working against a decaying signal, the longer someone’s been gone, the less a single email is likely to bring them back. Use AI to help build the actual sequence timing, not just individual emails.
A reasonable default structure: a check-in email at 60-90 days (no discount, just a genuine nudge), a stronger offer at 120 days if there’s still no response, and a final, honest “is this still a fit” email around 180 days that includes a real, one-click opt-out. SMS can meaningfully lift response on the final message specifically, since it’s often the channel someone hasn’t already learned to tune out the way they may have with email.
Ask AI to draft all three touches at once, with explicit instructions to vary tone and urgency across them rather than just changing the discount percentage each time, which is the laziest and most common version of a win-back sequence and the one customers recognize and ignore fastest.
Build in a stop condition, too. If someone clicks through and browses but doesn’t purchase, that’s a different signal than not opening at all, and it’s worth a lighter-touch follow-up rather than jumping straight to the next scheduled discount. Ask AI to help you define these branch conditions up front, based on whatever engagement data your email platform actually tracks, so the sequence adapts instead of blindly running the same three emails regardless of how someone responds along the way.
Here’s my honest take after building enough of these for clients. The most common mistake isn’t a bad AI-written email, it’s using AI to escalate the discount automatically every time someone doesn’t respond, training your most price-sensitive customers to simply wait out your sequence for the biggest possible offer. That’s not reactivation, it’s teaching people to ignore you until you pay them to come back.
The second mistake is skipping the segmentation step entirely and going straight to AI-generated copy for one generic blast, which defeats the entire point. A perfectly written email sent to the wrong segment with the wrong offer performs worse than a mediocre email sent to the right one.
And the one that actually damages your brand long-term: never building in a real, respected opt-out path. If someone’s ignored three win-back attempts, that’s a clear signal, not an invitation to email harder. Use AI to identify that pattern too, and suppress the segment instead of escalating to them again next quarter.
There’s a deliverability angle to this too, one that gets ignored until it becomes a real problem. Repeatedly emailing a segment that never opens anything drags down your sender reputation for everyone else on your list, not just the lapsed group. Treat suppression as protecting your active customers’ inbox placement, not as giving up on the dormant ones. A smaller, more responsive list consistently outperforms a larger, mostly-ignored one.
The DTC client from the opening example ended up suppressing about 4,000 customers entirely after the sequence ran its course, people who hadn’t opened, clicked, or responded to any of the three touches. That felt counterintuitive to her at first, actively shrinking the list she’d get remarketed to. Her deliverability metrics improved within a month, and the smaller, more engaged remainder of that list has outperformed the old, bloated one on every campaign since.
Automated win-back sequences typically reactivate 12 to 18% of lapsed customers, and Klaviyo’s own benchmark data across roughly 183,000 brand accounts puts average win-back flow revenue at $0.84 per recipient. Reactivating a lapsed customer is also estimated at 5 to 7 times cheaper than acquiring a new one, which is why win-back is usually underpriced revenue most brands aren’t systematically working.
Export your lapsed customers with order count, total spend, last product purchased, and days since last order, then ask AI to group them into segments based on likely reactivation approach, not just time since last purchase. Distinguish high-value lapsed customers from one-time-only buyers from recently-lapsed versus long-dormant, since each needs a genuinely different message, not the same discount at different intervals.
No, and leading with one is often a mistake. For recently-lapsed or high-value customers, referencing what they actually bought and giving a specific reason to return often outperforms an immediate discount. Save the stronger offer for later touches in the sequence, and always include an honest opt-out rather than escalating the discount indefinitely to people who keep ignoring you.
A reasonable default is three touches: a no-discount check-in around 60-90 days, a stronger offer around 120 days if there’s no response, and a final honest email around 180 days with a real opt-out option. Continuing to email lapsed, unresponsive customers indefinitely mostly damages sender reputation without meaningfully improving reactivation.
SMS tends to add meaningful lift specifically on later touches in a win-back sequence, since it’s often a channel the lapsed customer hasn’t already learned to ignore the way they may have with email. It works best as part of a multi-channel sequence rather than a replacement for email throughout.
This article draws on Klaviyo’s own published win-back flow benchmark data (based on roughly 183,000 brand accounts) and aggregated ecommerce retention research on reactivation rates and acquisition-cost comparisons, all fetched and verified live during this writing session on August 6, 2026.
Weekly tips. Real examples. Practical help for busy professionals.
We care about your data, check out our privacy policy.