Loyalty at your venue: what's proven to work and what does harm

If you've ever sat through a loyalty program pitch, you've seen this number: “program members spend 25% more.” Or 40%. Or twice as much.

It's true. And it proves nothing.

Because they compare program members with non-members. And the people who sign up aren't random: they're the ones who already come more often. Someone who drops in twice a year doesn't fill in a form at the till. The regular who would have come for the tenth time anyway does. The program didn't make them come more often; it registered them.

This is called selection bias, and it's in almost every number sold to venues. The right comparison looks different: take similar guests, give the program to some and not to others, and look at the difference. Such studies exist. They're far less encouraging.

We gathered six lines of research, including the ones that don't flatter trendy mechanics and don't flatter us. Here's what follows from them.

A caveat up front

All the numbers below come from other markets, other programs and other businesses: Japanese retail, American coffee shops, hotel chains. We're not claiming your venue will see the same.

That's exactly why the main takeaway of this article isn't “use this mechanic” but “measure your own.” How to do that honestly is at the end.

Birthday rewards: the most popular mechanic with a negative result

Nishio and Hoshino (2024) analyzed 210,657 members of the MUJI program. They didn't compare “members with non-members,” but people who received a birthday reward with a control group.

Customer lifetime value (CLV) among those who got the birthday reward: $72. In the control group: $125.

Not a smaller gain. A worse result.

The authors' explanation: the mechanic attracts impulse buyers who come for the gift and don't return, while for the most loyal guests it simply gives away margin at a moment when they would have come anyway.

It's one of the most popular mechanics in the industry. It gets switched on by default because it's nice and easy to set up. We support it too. But that's not a reason to turn it on without thinking; it's a reason to measure it separately before you scale it.

Stamps work. Just not the way you're promised

Here the evidence is strongest, and it’s on the side of the mechanic.

Nunes and Drèze (2006) compared two cards. The first had 8 empty slots. The second had 10 slots, with two already stamped “as a gift.” The effort was identical: in both cases the guest needed to make 8 purchases.

Completed the card: 34% versus 19%.

Almost twice as many, and the difference came from framing alone. It's easier to finish a journey you've started than to start from zero.

Now the part pitch decks leave out. Kivetz and colleagues (2006) followed 948 members of a coffee shop program and recorded a post-reward reset: after getting the reward, engagement drops below the original level. The guest relaxes, because the goal has been reached.

So stamps don't give you a growth curve; they give you a sawtooth. Ramp up, peak, dip, ramp up again. The mechanic works and it's useful, but the phrase “stamps increase loyalty” is inaccurate. They increase frequency within a cycle and cause a dip after it. That's fine if you plan for it. And it's an unpleasant surprise if you were sold a chart that only goes up.

Tiers: the mechanic that hits your best guests

A study in Marketing Science (2016) of 95,532 customers tested what happens when people are given a goal tied to status.

20% reached the goal. 80% failed.

And here's the key part: failing significantly reduced purchases after the campaign, and the sharpest drop was among customers with the highest status. The very people tiers are launched for.

The logic is clear: the higher you've climbed, the more it hurts to fall short. A guest who was “gold” for two years and then didn't make the threshold gets not motivation but an unpleasant message about their own failure.

In hotels and airlines, tiers make sense: the prize is big, status has real social value, people show it off. At a coffee shop, “Silver level” has no social value at all, while the failure mechanic works at full strength.

Tiers should be optional, not the default. If you turn them on, do it for a specific economic reason, not because they're in the interface.

The effect disappears where everyone has a program

A study of 358 chains in 27 countries shows something uncomfortable: in categories where every player has a program, the effect of any single program tends toward zero. It isn't an advantage, it's parity. A fee for the right not to lose.

Separately, work by Liu (2007): the effect of a program on frequent guests is practically zero. They were coming anyway. For them, a reward is mostly margin given away.

Together, these two findings mean the loyalty budget is often spent in the worst possible place: on people who would come without it, in a category where everyone already does the same.

Where to put it instead is below.

Gamification: badges, leaderboards, prize wheels

The classic work by Deci, Koestner and Ryan (1999), a meta-analysis that is still a core text in the psychology of motivation: tangible rewards significantly undermine intrinsic motivation. The effect is called overjustification.

The mechanism is simple and unpleasant. Someone came to your venue because they like it there. You started paying them in points for the same thing. The motive gets swapped: now they come for the points. Take the points away, and the original reason has already been weakened.

A badge doesn't give a sense of competence, a leaderboard doesn't give a sense of belonging, and both reduce the sense of autonomy. What's left is pure external motivation: maximum risk for minimum benefit.

Gamification looks harmless, because “it's just for fun.” But it works on exactly the resource a good venue already has for free.

The main thing: get a newcomer to their fourth visit

This is the most useful finding of the whole review, and it turns priorities upside down.

According to industry panel data (Paytronix, 2026), 95% of guests return after their fourth visit. Chipotle builds its model around a similar benchmark: three visits in the first 90 days.

In other words, loyalty is decided somewhere other than where it's usually bought.

A guest who has been to you four times is already yours. They don't need a card, points or a tier; they just keep coming. A guest who has been once is an open question, and that's exactly where every interaction carries the most weight. Yet most programs are built the other way round: the biggest rewards go to people who have already reached their tenth visit.

The practical takeaway for an owner: look at how many people get from their first visit to their fourth. If you don't know, that's the most expensive blind spot in your analytics. Not “how many program members do we have,” but “how many newcomers made it to a fourth visit.”

The budget works harder not on retaining the loyal, but on moving a newcomer from the first visit to the fourth. That's the best ROI the available evidence points to.

Non-monetary rewards beat money

Four independent studies point the same way.

Melnyk and Bijmolt (2015): monetary rewards had a statistically insignificant effect. Liu-Thompkins and colleagues (2022), a meta-analysis of 1,908 effect sizes: emotional drivers are about 50% stronger than rational ones. Wallström and colleagues (2024): cash rewards train transactional behavior, meaning the guest learns to come for the discount. The Bond industry report (2025): for the first time, access and experiences overtook financial rewards in guests' own ratings.

In practice it looks like this: the Thanx platform reports the average discount falling from 10% to 2.4% while engagement grew at the same time. Money was replaced with access and recognition.

For a venue, this means “a table without the queue,” “try a new menu item first,” “an invitation to a private tasting” and a simple “good to see you, the usual?” are often stronger than 10% off. And cheaper.

Reactivation: you have to act before the guest leaves

One more finding that changes the order of things. Working 60-90 days before a predicted lapse gives a higher save rate than any attempt after the guest has already gone.

It's logical, but almost nobody does it, because it requires predicting rather than reacting.

And the escalation should be a ladder: recognition first, then something of small value, and only at the end a discount. Not because we're protecting margin, but because a discount immediately teaches your base to wait for discounts. The same transactional reflex from the previous section.

What we're doing about it

Now, honestly, about our side, with the same degree of skepticism.

Stamps target specific menu items. Not “any purchase,” but chosen products. The mechanic is idempotent: the POS retries webhooks up to 16 times, and none of those retries adds a second stamp.

Notifications about points, tiers and stamps arrive in Telegram and in the app. The guest sees where they stand instead of guessing. And they're throttled: someone who buys coffee three times a day won't get three messages, because the fastest way to lose a channel is to flood it with trivia.

A control group is built into the plan as the next step. We'll say it plainly, because it's the same question this article started with: without a holdout, any report of ours would show “members spend more,” and that would be the same selection bias. We don't want to publish a number we've just criticized, so the measurement is being built at the same time as the impact report, not after it.

POS integration (Poster, Syrve, iiko): points come from the real receipt, not from someone's word at the till.

Reactivation through Telegram. Here we have a structural advantage, and it's purely economic. When a channel is paid, every contact costs money, so vendors send a discount straight away to pay for the message. For comparison, Airship charges 3.5p per SMS. Our channel is free, so escalation can start with recognition and small value, leaving the discount for the end of the ladder. That is, in exactly the order that works better.

What we don't promise

We won't tell you the percentage by which your revenue will grow. Nobody who hasn't seen your data knows it, and any such number in a pitch is either someone else's case study or the same selection bias.

We won't say that loyalty on its own will bring guests back. The study of 358 chains says outright that in a saturated category a program is more a condition of playing than an advantage.

And we won't sell a mechanic the evidence is against. We do have birthday rewards, because venues that want them need them. But we write about the MUJI study on our own blog instead of hiding it.

Where to start if you're choosing right now

Three questions worth answering before you launch anything.

1. How many of your newcomers get to a fourth visit? If you don't know, start there, not with choosing a mechanic. This number decides where the budget goes.
2. How will you tell the program's effect apart from selection? If there's no answer, any report six months from now will be pleasant and untrue. The control group is set up at the same time as the program; later it won't be possible. This applies to us too: that's exactly why we don't show an impact report until there's a holdout next to it.
3. What can you offer besides a discount? Access, attention, the first taste of a new menu item, a table at peak hour. According to the research these things are stronger, and they cost less.

And one general principle that conveniently comes from two different directions. Behavioral science says a mechanic open to everyone, with the same rules for all, works better than a personal gift. Ukrainian tax law says roughly the same thing: a non-individual discount available to everyone on equal terms isn't part of the guest's taxable income, while an individually assigned one becomes an additional benefit. A rare case where the more cautious option is also the simpler one.

Sources mentioned in the text: Nishio & Hoshino (2024, n=210,657); Nunes & Drèze (2006); Kivetz et al. (2006, n=948); Marketing Science (2016, n=95,532); Liu (2007); a study of 358 chains in 27 countries; Deci, Koestner & Ryan (1999, meta-analysis); Melnyk & Bijmolt (2015); Liu-Thompkins et al. (2022, 1,908 effect sizes); Wallström et al. (2024); Bond Loyalty Report (2025); Paytronix panel data (2026).