Designing Rewards That Actually Change Customer Behaviour

A car wash in the United States once ran two versions of the same loyalty card. Both required customers to collect enough stamps to earn a free wash. One card needed eight stamps, with none given away. The other needed ten stamps — but two were already stamped when the customer received it. Mathematically, both cards demanded the same six additional purchases. Psychologically, they were nothing alike. Customers holding the ten-stamp card with a head start finished it faster and came back more often than those working the "honest" eight-stamp version.

That finding comes from a 2006 study by Ran Kivetz, Oleg Urminsky and Yuhuang Zheng, "The Goal-Gradient Hypothesis Resurrected: Purchase Acceleration, Illusionary Goal Progress, and Customer Retention," published in the Journal of Marketing Research. It is one of the cleanest demonstrations in the loyalty literature of a truth most reward programmes ignore: the way a reward is framed changes behaviour more than the size of the reward itself. Give people two free stamps and a longer card, and you have not spent a cent more — you have just made the finish line feel closer.

This is the central argument of this piece: most loyalty programmes are accounting exercises dressed up as behavioural ones. They pay people for the behaviour they already had. A rewards architecture that actually changes behaviour looks different — it uses the structure of the reward, not just its value, to move customers toward the next purchase, the next visit, the next habit. Get the structure wrong and you can double the points and still move nothing.

Why do most loyalty programmes fail to change behavior?

Most loyalty programmes fail to change behaviour because they are designed to reward it retroactively rather than shape it prospectively. A points-per-dollar scheme simply reimburses spend that would have happened anyway — it is a discount with better marketing. It does nothing to alter the customer's decision at the moment that matters: whether to choose you over the alternative, whether to come back this week rather than next month, whether to try the higher-margin product instead of the familiar one.

Frederick Reichheld's influential 2003 Harvard Business Review piece, "The One Number You Need to Grow," made the case that loyalty is fundamentally about the willingness to recommend, not the accumulation of points. Points programmes can coexist happily with declining loyalty because they measure and reward transactions, not the relationship. A customer can be enrolled, engaged with the app, and quietly shopping a competitor for anything that isn't on promotion.

The tell is in the redemption data most brands never publish: a large share of points issued are never redeemed at all, and of those redeemed, a large share go to customers who were already the most frequent, highest-spending segment before the programme existed. You are not buying new behaviour. You are subsidising behaviour you already owned — a loyalty tax paid to people who never asked to be taxed.

What's the difference between reinforcing loyalty and manufacturing it?

Reinforcement rewards behaviour that already exists; manufactured loyalty creates behaviour that would not otherwise happen. The distinction matters because they require entirely different reward mechanics, and most programme designers only know how to build the first kind.

Reinforcement says: thank you for the ten trips you were always going to make — here is 5% back. It is generous, predictable, and behaviourally inert. Manufactured loyalty says: you are two purchases from a benefit you can see, feel, and almost taste — and it uses the machinery of progress, scarcity, and framing to pull the customer across a line they had not planned to cross. Airlines have understood this for decades. The tier just above where a flyer currently sits is rarely a rounding error away by accident; airlines design status thresholds so that frequent flyers can see the next tier and feel the pull of finishing it, a manufactured incentive that has kept status programmes central to airline economics since American Airlines launched AAdvantage, the first modern frequent-flyer programme, in 1981.

The behavioural lever underneath both the car wash card and the status tier is the same: the goal-gradient effect, first documented in animal behaviour by psychologist Clark Hull in the 1930s and revived for consumer contexts by Kivetz, Urminsky and Zheng. Effort and motivation intensify as the perceived distance to a goal shrinks — not because the reward has changed, but because the finish line has moved into view. A rewards programme that never lets customers see the finish line is leaving its single cheapest behavioural lever unused.

How does loss aversion make expiring points more powerful than earning points?

Loss aversion makes expiring points and status more powerful than earning new ones because losing something you already hold is felt roughly twice as intensely as gaining an equivalent amount, a finding that traces back to Daniel Kahneman and Amos Tversky's prospect theory and was tested directly in the context of ownership by Daniel Kahneman, Jack Knetsch and Richard Thaler in their 1990 study "Experimental Tests of the Endowment Effect and the Coase Theorem," published in the Journal of Political Economy. Once a customer holds status, miles, or points, those things are no longer an abstract future benefit — they are an asset already endowed to them psychologically, and the threat of losing that asset motivates action far more reliably than the promise of gaining a new one of equal size.

This is why the most behaviourally sharp elements of a loyalty programme are rarely the earning mechanics — they are the expiry rules, the status "requalification" periods, and the countdown emails that say a member is about to drop a tier. Renascence's work in customer loyalty strategy repeatedly finds that a well-timed "you're about to lose this" message outperforms a "you could earn this" message on almost every retention metric a client tracks, precisely because it activates loss aversion rather than mere anticipation.

Used carelessly, though, this same mechanic curdles into anxiety rather than motivation — a member who feels perpetually on the edge of losing status disengages rather than strains toward it. The craft is in calibrating the threat so it is real enough to matter and close enough to be solvable with one more visit, not a distant cliff edge that feels hopeless.

Can unpredictable rewards work, or does variability just feel unfair?

Unpredictable rewards can work, and often outperform fixed ones, provided the variability sits on top of a floor customers can trust — unpredictability alone, with no guaranteed baseline, reads as unfair rather than exciting. The psychological mechanism is the same one that makes slot machines and surprise mechanics in mobile games so sticky: a variable ratio of reinforcement produces more persistent behaviour than a fixed one, because the brain cannot predict which action will pay off, so it keeps trying.

Applied responsibly, this looks like a loyalty programme that guarantees a baseline reward for every qualifying purchase, then layers in occasional surprise multipliers, bonus draws, or "mystery" tier upgrades on top. The guarantee protects trust; the variability protects attention. Sephora's Beauty Insider programme and Starbucks Rewards both use this layering — predictable point accrual sitting underneath periodic bonus-point events and surprise perks — a structure that keeps the programme feeling alive rather than mechanical, without ever making members feel cheated on the basics.

Where this goes wrong is when brands strip out the guaranteed floor to fund the surprises — cutting base earn rates to pay for flashier promotions. Customers notice the base cut long before they appreciate the occasional bonus, and the net effect is resentment, not delight.

What does a reward architecture that actually changes behavior look like?

A reward architecture that changes behaviour is built around the specific action you want more of, not around a generic point currency that tries to serve every objective at once. The design sequence that tends to work follows a consistent pattern:

1. Name the single behaviour you are trying to shape. Not "increase loyalty" — a specific, observable action: switch from monthly to quarterly renewal, add a second product line, refer a friend within 30 days of joining. Vague objectives produce vague, universal rewards that move nothing.
2. Map where that behaviour currently breaks down. Use the actual customer journey — where does intent exist but action stall? This is friction-hunting territory, and it belongs in a proper journey mapping exercise before a single reward is designed, because a reward aimed at the wrong friction point is wasted spend.
3. Choose the mechanic that matches the psychology of that stall point. A visible, near-term goal calls for goal-gradient design (a progress bar, a card with a head start). A risk of attrition calls for loss framing (status at risk, points expiring). A habit you want to seed calls for variable reinforcement layered on a guaranteed floor.
4. Set the threshold where it is achievable but not trivial. Behavioural economics research on goal-setting consistently shows that goals perceived as just within reach generate more sustained effort than goals that are either too easy or clearly out of range — the sweet spot is a stretch, not a gift and not a wall.
5. Make progress visible at every touchpoint, not just at redemption. The reward has to be felt during the journey, not only announced at the end of it. This is where a points balance sitting in a forgotten email does less work than a progress indicator surfaced inside the app, at the till, or in-store.
6. Test the counterfactual before scaling. Run the new mechanic against a control group doing the old one. If behaviour among the treated group is not measurably different, you have built a discount, not a rewards architecture — go back to step one.

Programmes that skip straight to step six's opposite — scaling before testing — are the ones that quietly cost the most and prove the least. A well-run loyalty management platform earns its budget by making steps two through five operationally possible, not just theoretically nice.

Which reward mechanics tend to backfire?

Some reward mechanics reliably backfire because they trigger the wrong psychology or erode the trust the whole programme depends on. Watch for these specifically:

• Rewards for behaviour customers already do at 100% rate. Paying existing habitual buyers for purchases they were always going to make is pure margin erosion with zero behaviour change — the definition of the loyalty tax.
• Redemption friction that undermines the reward's own promise. A points system that requires ten clicks, a phone call, or a minimum threshold so high most members never reach it converts a goodwill gesture into a grievance — the sludge, in Richard Thaler's terminology, cancels out the intended nudge.
• Devaluing points retroactively. Changing redemption rates after customers have accumulated a balance activates loss aversion against the brand itself, and airline and hotel loyalty schemes that have done this have absorbed real reputational damage as a result.
• Rewarding the metric instead of the outcome. Incentivising app downloads, sign-ups, or check-ins without tying them to actual revenue-relevant behaviour produces vanity engagement that dies the moment the incentive stops.
• One-size-fits-all tiers in a business with wildly different customer segments. A frequent, low-basket grocery shopper and an occasional, high-basket one need different thresholds and different rewards; forcing both through the same ladder rewards neither well.

How do you know if rewards are actually changing behavior, not just being redeemed?

You know rewards are changing behaviour when treated customers do something measurably different from a comparable group that received no incentive — more frequent visits, a wider basket, a longer tenure, a faster return after a lapse — not merely when redemption rates look healthy. Redemption is a vanity metric on its own; it tells you the reward was claimed, not that it caused anything.

The more rigorous test is incremental lift: take a matched control group that behaves similarly to your target segment before the intervention, run the new mechanic on the treatment group only, and compare outcomes over a fixed window. If lifetime value, repeat-purchase rate, or churn among the treated group doesn't move beyond what you'd expect from natural variation, the mechanic isn't earning its cost — however good the redemption numbers look on a dashboard.

This is also where the finance conversation and the behavioural conversation have to sit at the same table. A reward that shifts behaviour but costs more than the incremental margin it generates is not a success story — it is a more sophisticated way of losing money. Modelling that trade-off properly, and putting a number on what a percentage-point improvement in retention is actually worth, is exactly the kind of exercise a CX ROI calculator is built for, and it should sit alongside any loyalty business case before a single point is issued.

What role does behavioral economics play beyond the reward mechanic itself?

Behavioural economics shapes not just which reward works, but how it is presented — the same reward, framed differently, produces different behaviour. Anchoring the value of a reward against a higher reference point ("worth $50" rather than "500 points") changes perceived generosity without changing cost. Defaults matter too: auto-enrolling customers into a loyalty tier at the point of purchase, rather than asking them to opt in later, captures far more members simply because the effortful choice has been removed — a lesson straight out of Richard Thaler and Cass Sunstein's work on choice architecture.

The IKEA effect is worth borrowing from as well. Customers who build something toward their reward — stamping a card, curating a personalised set of perks, choosing their own redemption path — value that reward more than one simply handed to them, because effort invested breeds attachment. This is why the most durable loyalty mechanics rarely feel purely transactional. Building this kind of thinking into a programme from the outset is properly the territory of applied behavioural economics work, not an afterthought bolted onto a marketing calendar.

None of this replaces the basics of good service. A rewards programme cannot manufacture loyalty toward a brand that consistently disappoints on the fundamentals — reliability, fairness, ease. Points can accelerate an existing relationship; they cannot invent one from nothing.

Designing for the next visit, not just the next redemption

The best test of any rewards programme is a simple one: if you removed it tomorrow, would anything about how customers behave actually change? For most points schemes, the honest answer is no — customers would grumble, then carry on exactly as before, because the programme was never shaping their choices in the first place. It was reimbursing them.

The programmes worth building are the ones where the answer is yes — where the progress bar, the near-miss tier, the loss-framed reminder, or the surprise bonus is quietly doing real work on real decisions, purchase by purchase. That is a harder programme to design and a much easier one to defend in front of a finance team, because its value shows up as incremental behaviour, not just issued points sitting unredeemed on a balance sheet. Design for the next decision, not the last transaction, and the loyalty tends to follow.