The usual answer is a day count. Ninety days of no opens, or six months, and then a goodbye email. That answer is not wrong so much as it is answering a different question, because the day count only tells you who qualifies. What decides whether a winback flow is worth running at all is how many people qualify in a normal week, and on most stores that number is far smaller than the flow's send volume suggests.
What happens the moment you turn the flow on
Klaviyo triggers a winback or sunset flow on segment membership, and it only fires for profiles that enter the segment after the flow exists. Beside the trigger card sits an "Add past profiles" control, and everybody building one of these clicks it, because the alternative is a flow that starts empty while the entire lapsed list sits outside it doing nothing.
That click is the event. A store with 40,000 profiles and a six-month-inactive segment holding a third of them will send more mail in the first hours of that flow's life than it sends in a normal month. Everything after is a trickle: the people who cross the inactivity line this week, which is a function of how many addresses you collected six months ago and how many of them went quiet.
So the flow has two completely different periods, and they get averaged together in every report anyone looks at. The first is a one-off bulk send to your worst addresses. The second is the actual ongoing flow, and it is small.
The account where this showed up as a number
We took over a sports-supplement brand running two re-engagement flows. Across the window we measured, those flows sent 64,342 emails. Revenue attributed to them: €0. Unsubscribes: 635.
The instinct is to rewrite the copy. The copy was ordinary, and it was not what produced that result. The segments those flows mailed did not grow. The backfill went out, the pool was spent, and what was left was a fixed audience that had already been asked and had already declined. The flows were not underperforming against some benchmark. They had finished, months earlier, and nothing in Klaviyo announces that.
635 unsubscribes is the part worth sitting with. That is 635 people who were on the list, costing nothing much, who were prompted into a decision and made it. Some of them would have bought again on a campaign eighteen months later. A winback flow aimed at a dead pool does not just fail to earn, it spends something.
The entrant rate is the decision
Open the segment and look at how many profiles joined it in the last 30 days, ignoring the backfill. That number is your flow's real audience per month.
If it is a few hundred, you have a flow. Build it properly, leave it running, and it will quietly recover people for years. If it is eleven, you do not have a flow, you have a cleanup task that runs twice a year, and the automation is costing you more attention than the outcome justifies.
The window feeding that segment has to be scaled to how often you send, for the same reason an engaged segment cannot be a flat 30 days. Somebody who receives two emails a month has had two chances to show a sign of life inside 90 days. Calling them lapsed on that evidence is a calendar artefact, not a finding about them.
This is also why winback sits where it does in the build order. Welcome, abandoned cart and post-purchase all mail people whose interest is current, and the order to build flows in puts them first for that reason. Winback is the flow that asks the coldest people on your list for something, which makes it the one most likely to be built enthusiastically and left running past its usefulness.
When to actually stop
Stop when the person has had a fair chance to answer and has not: the window scaled to your cadence, then three emails, then a grace period long enough that one of those emails landed in a week they were reading mail. Klaviyo's own guidance caps the flow at three emails and recommends a delay at the end before you act, stretched to seven or ten days if you send infrequently.
Add the two flow filters that make the sequence honest, checking that the profile has not opened and has not clicked since starting the flow. Anyone who answers drops out before the goodbye email, which is the difference between a sunset sequence and a machine that unsubscribes your re-engaged customers.
Then suppress rather than delete. A suppressed profile stops costing you money and stops receiving mail, and it keeps its order history, so the person can come back on their own and be recognised. That is the same pass that cuts an active profile bill without losing buyers, and it is worth running whether or not you ever build the flow.
How many emails should a winback flow have?
Three, and three is a ceiling rather than a target. The people in this flow have already ignored everything else you sent. A fourth and fifth attempt raises complaints against an audience that generates almost no revenue, and complaints are scored against the domain that sends every other email you have.
Two is defensible if your list is small. One is a goodbye email, not a winback flow, and it skips the part where you actually try.
Should the last email carry a discount?
The first one should not. Lead with the discount and you have taught your list that going quiet is how you get 20% off, and the people most alert to that lesson are the ones who were about to buy anyway.
By the final email the calculus changes, because the alternative is suppression. A real offer in the last message, framed as a last attempt, is the one place a deep discount costs you nothing you were otherwise going to keep.
Does suppressing someone lose the sale later?
No. Suppression removes a profile from sends and from billing, and leaves everything else intact. They can still buy, still resubscribe through a form, still trigger flows if they come back through the front door.
What you give up is the ability to email them unprompted, which is the thing they have spent six months declining to read.
