Loyalty Marketing: The Program Is Not the Strategy

The dashboard is open on a Tuesday, and the line your loyalty marketing program was built to move is flat. Redemptions climbed again last quarter and repeat purchase sits where it sat a year ago. Nobody says that out loud, so the conversation turns to tiers and a new partner. Only the logo on the login screen changes. So: if the reward vanished tomorrow, how many of those members would still come back without being paid to? The answer is already in your tables.

Somewhere in your member file is a group of people who joined, never redeemed a point, and kept buying at the rate they always did. Deloitte's 2025 survey of 5,564 US members found the average consumer enrolled in eight programs and active in five, with 51 percent engaging with only one. Bond's 2025 report put it at 17.4 enrolled and 8.8 active. Antavo's 2026 report found 74 percent quiet within two months while 3.4 percent formally cancel. A member file is a mailing list until someone proves otherwise, and you have probably reported it as growth. So the reward pays a margin cost for customers it never moved.

What a Loyalty Program Actually Buys

A program buys three things you can observe. An identity, because you know who bought. A permission, because a member agreed to hear from you. A nudge inside a purchase cycle that already exists. What it does not buy is a preference that was not there, a defense against a better product, or forgiveness for bad service.

Leenheer, van Heerde, Bijmolt and Smidts measured that distance in the International Journal of Research in Marketing in 2007. Their Dutch household panel covered all seven grocery loyalty programs. Members held 36 percent of their wallet with the sponsor, non-members 7 percent. Modeling the join decision, since loyal households enroll more readily, cut the effect to 4.1 points. About seven eighths of the visible gap was selection. Loyal customers join, and joining did not make them loyal.

A case study reporting members who spend twice as much is publishing the uncorrected number.

Kivetz, Urminsky and Zheng followed 948 members of a cafe card program in the Journal of Marketing Research in 2006 and found purchases accelerating roughly 20 percent as the card neared a reward, then falling sharply after redemption. A program changes what a member does next. It rarely changes why a customer came.

Reading the Numbers Honestly

Starbucks is the cleanest public case. In its fiscal 2025 first quarter, North American comparable transactions fell 8 percent while US 90-day active Rewards members rose 1 percent to 34.6 million.

Why a rising redemption rate can be a warning

Redemption rate is rewards redeemed divided by rewards issued. It rises when the reward gets easier to reach, or when you lowered the threshold. On its own it measures supply, not demand.

The reading that matters is the pair. Redemption rising while repeat purchase stays flat means more rewards claimed and no second purchase produced, and in a quarterly review that line reads like good news.

Composition makes it worse. Points accrue on spend, so redemption concentrates in the heaviest buyers. Taylor and Neslin studied a supermarket program in 2005 and found only about 20 percent of members redeemed at all. Storewide sales still rose 6 percent during the campaign and 17.5 percent after redemption.

The breakage illusion

Breakage is revenue recognized from value that will never be redeemed. Starbucks booked USD 200.4 million in company-operated store revenues and USD 22.0 million in licensed store revenues for the year ended 28 September 2025, against a deferred balance of USD 1,751.7 million. It runs a mature program with low breakage, which makes the filing instructive.

Value a member earned and never spent becomes revenue this period, which flatters the return in the year it happens. Antavo found 26.2 percent of points unspent and 11.9 percent expiring in 2026. Breakage then reads as a count of members who found the reward not worth the trip.

How to test whether the reward did the work

Reward usage against non-usage. Split members by whether they redeemed and compare repeat purchase over 90 days. This is not causal, but it kills the naive reading.

Time to second purchase. Measure days from first to second order, and compare with customers acquired six months before launch. A working program shortens the interval between the two, and a rebate does not.

Margin after redemption. Add the reward cost to redeemed orders, compute gross profit, compare with member orders that did not redeem. Lower margin with no earlier return means buying revenue at a loss.

A holdout, the only version that settles it. Withhold the program from a randomly assigned slice for one full purchase cycle, at least 90 days, then compare repeat rate and gross profit against the invited group. The slice must detect a five point difference.

Only the last one produces a number you can defend, because it is a revenue attribution problem before it is a loyalty one. The first three tell you whether a holdout is worth the setup cost, and each takes an afternoon.

The number worth computing first is the share of last quarter's redemptions that went to customers who would have bought anyway. A holdout gives it directly. Without one, the nearest proxy is the repeat rate of redeemers before they first redeemed. Both belong beside the metrics you already review, as our guide to marketing KPIs argues.

Loyalty Problem or Product Problem?

If the only per customer record is the reward ledger, the only explanation for a quiet member is an unattractive reward. That explanation comes from the record's shape, not from the customer.

1. Repeat rate by product concentration. If repeat sits in one SKU, you are looking at a consumption cycle. Replenishment goods repeat with no program at all.
2. Repeat rate against recommendation intent. Morgan and Rego tested six satisfaction and loyalty metrics against future performance in the Journal of Marketing in 2006, and net promoter measures were not superior predictors.
3. Redemption trend against repeat trend, same period, same axis. Divergence is the signal that opens this article.
4. Redemption share by prior spend decile. If the top decile claims most rewards, the program is a rebate for the already loyal.
5. Complaints, returns and delivery failures ahead of the lapse. Departure without a complaint is common, so look for friction that never became a ticket.
6. Discount depth as a share of revenue. If the cost is mostly reward value and the reward is a discount, you are running a price cut with a membership layer.

Ulta Beauty reported more than 46 million members and about 95 percent of sales from members in its fiscal 2025 10-K. Chewy reported Autoship sales of USD 10,497.1 million, 83.3 percent of net sales. At 95 percent penetration the program is the customer account, and at 83 percent subscription revenue the mechanism is a delivery commitment. Assortment and replenishment do the work, and no non-member control is left.

When Rewards Are the Wrong Instrument

The category's premise is that retention is worth a great deal: a five point gain lifts profit by 25 to 95 percent, attributed to Bain and repeated by Harvard Business Review in 2014. It describes the value of retention, not your program.

Start with a product that repeats on a fixed cycle. The second purchase is a property of consumption there, so a reward on top discounts a decision already made, and the customer was going to make it anyway.

Service or availability is the second case. A reward cannot compensate for a delivery that arrives late, and when people recommend you and still do not return, the gap sits in fulfillment.

A base too small to carry the operating layer is the third. The arithmetic fails before the design does. Integration, reward liability and reporting are fixed costs that do not shrink with the member file, so the cost per member can exceed what the behavior is worth.

The fourth is a thin margin against a discounted reward. Then the program is a price cut with a login. It produces redemption and revenue, not the profit the business case promised.

If You Build One Anyway

Three choices decide whether it survives year two.

The first is what the reward is priced against. Points on spend reward the customers who were already spending, which is the selection effect with a budget attached. Points on a behavior you want more of, such as a first repeat or a second category, price something that does not exist yet.

The second is whether the threshold is reachable inside one purchase cycle. A member who needs three cycles to reach the first reward gets nothing in the cycle where the habit forms, and Antavo's 74 percent quiet within two months says they will not wait. A quiet member is not churned, only unreachable.

The third is what happens the day after redemption. Kivetz and his co-authors found pace dropping sharply once the reward was claimed, the predictable behavior of a closed goal. A program that resets the member to zero asks them to start from nothing. One that opens the next threshold immediately keeps the goal ahead of them.

Conclusion

A loyalty program is a mechanism for paying people to do something. It works when the something is new, when it is reachable, and when you can prove it would not have happened anyway. It fails quietly when it pays people for behavior they already had, and the failure looks like success on the redemption chart for years.

The strategy is knowing which customers came for the product and which came for the reward, and that is where the budget question gets decided.

If you had to defend the program's budget next quarter with one number that is not a redemption rate, which number would you bring? For teams that already treat the reward as one lever rather than the whole answer, customer advocacy is the harder question.