Loyalty · Testing

Does a win-back discount earn more than a full-price reminder?

By Mirow · October 27, 2026 · 6 min read

Judge a win-back discount by the extra contribution it produces per eligible customer, after discounting the orders you would have received anyway. Compare a discounted message with the same reminder at full price. Add a no-message group if you also need to know whether the reminder itself earns its cost.

A coupon can bring back more buyers while leaving less money for the business. The useful question is how much the offer adds beyond a cheaper alternative you could actually run.

Separate the reminder from the discount

Choose one group of customers whose purchase history makes a win-back message reasonable. Define the product category, time since purchase and exclusions before assigning anyone. A customer who buys furniture every few years is not necessarily overdue after 90 days. If the decision is when a consumable should run out, start with the replenishment timing guide.

Randomly assign eligible customers to three groups before sending. Keep each customer's assignment fixed through the measurement period:

• A: no win-back message. This group shows what happens under the remaining ordinary marketing.
• B: a reminder without a discount. Use the same product, send time and basic message as C.
• C: the reminder with a discount. Change only the offer and the wording needed to explain it.

C minus B estimates the effect of adding the discount offer. C minus A measures the whole discounted-message package. B minus A measures the reminder. A two-group discount-versus-no-message test cannot separate a useful reminder from an unnecessary discount.

Klaviyo supports random splits in flows. Use a fixed assignment record and confirm the resulting group counts; a random split does not guarantee equal counts. Keep people from entering another arm on a later trigger.

More buyers, less contribution

This is a hypothetical 30-day example in USD, unrelated to Mirow client results. Each group has 1,000 assigned customers. Every returning buyer places one order for the same $80 basket. The discounted group gets 15% off, or $12 per order, so its net sales are $68 per order.

Assume $40 of variable cost per order in each group: product cost, payment fees, packaging, fulfilment and business-paid shipping. Holding that total constant is a simplifying assumption; use the actual costs for each basket in a real test. There are no returns or customer-paid shipping charges in this example, and sales tax is excluded. Each email group costs $20 more to send than the no-message group. Other marketing is held comparable; its common costs are outside this comparison.

Scroll horizontally to see all columns.

Contribution here remains before fixed overhead, agency fees, creative production and other marketing costs. Any additional cost caused by the offer belongs in the decision before rollout. Discounts already reduce net sales; do not subtract them a second time.

The discounted message leaves $200 more than no message, but $180 less than the full-price reminder. It produces 15 more orders and $420 more net sales than B. Those gains still fail to cover the contribution given up through the discount.

The arithmetic is easier to see as two pieces. The 15 additional orders would contribute $600 at the full-price margin: 15 × ($80 − $40). Discounting all 65 orders costs $780: 65 × $12. The result is $600 − $780 = −$180. Email costs cancel in the C-versus-B comparison because both cost $20.

This is a decomposition of hypothetical group totals. It does not identify which individual buyers would have returned without the offer, and it is not evidence of a statistically reliable winner.

Set the economic hurdle before the send

Under these assumptions, each discounted order leaves $28 before email cost. To exceed B's $1,980, C needs:

Discounted orders × $28 − $20 > $1,980

That means at least 72 orders from the 1,000 assigned customers. At 72, contribution is $1,996, only $16 ahead of the reminder. The required order increase is 44% over B's 50 orders, while the example delivers 30%. Seventy-two is an economic break-even hurdle for this scenario, not a sample-size recommendation or a statistical stopping rule.

If the offer still produces only 65 orders, a $9 discount would leave $1,995 after email cost: 65 × ($80 − $9 − $40) − $20. That is just $15 ahead of B. A smaller discount may change response, so this supplies a new test hypothesis, not a promise that the cheaper offer will work.

Product mix, shipping thresholds and returns can move the hurdle. Calculate contribution from each customer's actual orders when baskets differ. The break-even contribution guide gives the broader cost definitions.

Count every assigned customer

For each arm, sum net sales after discounts and refunds, subtract variable order costs and additional campaign costs, then divide by all customers originally assigned to that arm. Include customers who do not open, click or buy. Honor changed permissions and suppressions, but retain those customers in the assigned-group analysis rather than quietly removing them.

Use all store orders you can reliably match to those assigned customers during an equal elapsed window from each customer's assignment, including orders without the coupon and orders credited to another channel. Reconcile customer identifiers and report unmatched orders as a measurement limitation. If an offer changes which product someone buys, the total customer contribution must capture that substitution.

Klaviyo's attribution model assigns revenue to eligible message interactions within configured windows. That answers a different question from whether the message caused a purchase. Coupon redemption and attributed revenue are useful diagnostics; neither supplies the no-offer comparison.

If the groups have different assigned counts, compare contribution per assigned customer instead of raw totals. For a planning estimate at C's group size, multiply the C-minus-B per-customer difference by C's assigned count. Keep the uncertainty around that difference visible.

Write a test card before building the flow

1. Population and assignment. Write the inactivity definition, eligible products, permission requirements and exclusions. Freeze assignments before the send. Prevent a second win-back flow from treating the same customer differently.
2. Offer boundaries. Specify the discount, expiry, eligible products and whether it combines with other offers. Use recipient-restricted codes where supported, and check for leakage into the other groups.
3. Observation and costs. Choose the primary window and a later follow-up before launch. Give every assigned customer the full elapsed follow-up before comparing groups. Thirty days is illustrative here. Allow for the category's buying cycle and a defined returns-reconciliation period. A purchase moved from next month into this month may disappear as a gain over the longer window.
4. Decision and precision. Set the minimum additional contribution worth implementing, then plan sample size around the variation in customer contribution and the precision needed to distinguish that amount. The 1,000-person groups above are invented for arithmetic. Avoid adding arms if the audience cannot support a useful comparison; test B versus C if the discount is the decision.
5. Operating checks. Keep ordinary marketing comparable, record concurrent promotions, and name the owner and review date. Track unsubscribes per delivered email and complaints using a stated denominator. Do not stop when one arm first looks ahead; apply the planned analysis and operational stop rules.

If the measured contribution difference is too uncertain to distinguish a useful gain from a loss, keep the result inconclusive. Start with one eligible segment and one offer decision. Before building the coupon, calculate how many extra orders it needs to beat the full-price reminder.