In brief
A loyalty program either changes what people do or discounts what they were already doing. Most of the argument about whether they work is really an argument about which of those is happening, and you can settle it with your own numbers in ninety days.
The short answer
Sometimes, and the variable is not the software. A loyalty program works when it changes behaviour — an extra visit, a larger basket, a customer who would otherwise have drifted. It fails when it pays a discount to people who were coming anyway. Both look identical from behind the counter, and telling them apart is the only question worth asking.
Nobody can answer it for you with a number from someone else’s business. What follows is how to answer it with yours.
The subsidy problem
Here is the mechanism that decides everything.
You run a ten-stamp card. A hundred customers join. Forty-three rewards get claimed a month. Every one of those rewards costs you the cost of goods, and every one is real money.
Now split those hundred customers into two groups. The ones who came in weekly before the card existed and still come in weekly are being subsidised — you are handing them a free coffee every ten weeks in exchange for behaviour you already had. The ones who came in fortnightly and now come weekly are the program working.
The program’s profit or loss is the margin from the second group minus the reward cost of both groups. That is the whole model. It is also why “do loyalty programs increase sales” has no general answer: it depends entirely on the ratio between those two groups in your particular shop.
The uncomfortable part: your best customers are always in the first group. A loyalty program is structurally a transfer from your margin to the people who liked you most already. It only pays if it moves enough of the second group to cover that transfer.
When a program provably does not work
Four business shapes where the arithmetic cannot close, and no vendor will tell you this:
Genuinely infrequent purchases. A mattress shop, a wedding photographer, a boiler service. If a customer buys once every three years, a ten-purchase card is a thirty-year commitment. There is nothing to be loyal to on a schedule anyone can hold.
Pure price competition. If customers pick by price every time and there is no switching cost, a loyalty reward is a delayed discount competing against an immediate one. The immediate one wins.
No repeat visit to capture. A business whose customers are tourists, or one-time referrals, or people who moved into the area and will move out.
A margin too thin to fund the reward. If your gross margin per visit times your threshold is smaller than about four times the reward cost, the program eats more than it can plausibly return. That is one calculation and it takes a minute. There is a worked version in what a loyalty program actually costs.
Three of those four are settled before you buy anything.
The disadvantages nobody puts on the pricing page
Margin leak on your best customers. Covered above. It is not a bug, it is the cost of the mechanism, and it should be in the budget rather than discovered later.
Staff friction. Every stamp is a moment where someone has to remember to ask, and a program that adds ten seconds to a queue at 8:15am will quietly stop being offered. The failure is invisible: enrolments just stop, and nobody reports it.
A price expectation you cannot take back. Once customers expect the tenth free, removing it reads as a price rise. Programs are much easier to start than to stop.
Data you are now responsible for. A list of customers is a data-protection obligation, not just an asset. Consent, retention, a way out. Small, but real, and it did not exist before.
The exit. Ask before you start: if you stop paying the vendor, what happens to the balances and the list? A program you cannot leave with your own data has a cost that never appears on an invoice.
The advantages that are real
To be fair to the mechanism, three things it does that a plain discount does not:
It pays in arrears. A standing 10% off is given to every customer on every visit, including the ones who needed no persuading. A card pays only after the visits that earned it. Same discount, better targeting, entirely because of the timing.
It creates a reason to come back that is not price. A card at seven of ten is a small sunk cost the customer carries. That is a weak force, but it is a force, and price cutting does not generate one.
It makes the invisible visible. Most local businesses genuinely do not know how many of their customers are regulars, how many stopped coming, or when. A readable ledger answers that whether or not the reward ever pays for itself. For some owners that alone is worth the nine pounds.
How to tell, with your own numbers, in ninety days
You cannot measure incrementality properly without a control group, and a single shop does not have one. But you can get close enough to make a decision.
| Day | Measure | What it means |
|---|---|---|
| 0 | Visit frequency of your named regulars, before launch | Your baseline. Write it down. Without this the rest is guesswork. |
| 30 | Enrolments per week | Below expectation means staff are not asking. Fix that before judging the offer. |
| 60 | Share of cards that reached a second stamp | The honest early signal. A card taken and never used again says the reward is not worth returning for. |
| 90 | Visit frequency of enrolled customers versus your day-0 baseline | The whole question. If it has not moved, the program is a subsidy. |
The day-0 number is the one people skip, and skipping it makes the day-90 number unreadable. You cannot detect a change from a number you never recorded.
Two honest caveats about that comparison. Enrolled customers are self-selected — the people who join a loyalty program are the ones who were already more engaged, so their frequency would look better than average even with no program at all. And seasonality moves visit rates on its own. Neither is a reason to skip the measurement; both are reasons not to over-read a small difference.
What to do with the answer
If frequency moved: the program works, and the next question is whether the threshold is set right. If it did not move but enrolment and activation are healthy: the reward is not worth changing behaviour for, and a bigger reward is the wrong fix before you have tried a nearer one. If enrolment never happened at all: you have a counter problem, not a loyalty problem, and no vendor can fix it.
If none of those apply and the arithmetic in the first section says your margin cannot fund it, the right answer is to not run one. That is a legitimate outcome, and it is cheaper to reach on paper than after a year.