Retention / Loyalty
How to Design an Ecommerce Loyalty Program
// reward the customers you already have before chasing new ones
By Fracto Solutions
October 30, 2026
The short answer
A loyalty program rewards customers for repeat purchases and engagement, turning one-time buyers into regulars. It works because repeat customers cost less to sell to and spend more over time. Design it to reward behavior you actually want, pick a structure (points, tiers, or cashback) that fits your margins and purchase frequency, and keep the rewards simple, attainable, and margin-aware.
34% v 19%
6.3c v 6.5c
~20%
Why run a loyalty program?
To increase how often and how much your existing customers buy, and to do it at a reward cost your margin can carry. Note what is missing from that sentence: the claim that loyal customers are cheaper to serve. That claim is standard in the industry and the best evidence does not support it.
We need to correct this page, because an earlier version led with it. Reinartz and Kumar examined the customer databases of four companies, more than 16,000 customers, across four years, and found no evidence that long-standing customers were cheaper to serve. At the mail-order company, communication cost per dollar of sales was 6.3 cents for long-standing customers against 6.5 cents for newer ones, a difference of no practical consequence. At the corporate service provider, the only strong correlation suggested loyal customers were more expensive to serve. And on price, long-term customers paid 5% to 7% less there, with regular customers paying 9% less in one mail-order category, leading the authors to conclude that “a loyal customer … is actually more price sensitive than an occasional one.” Longevity-to-profitability correlations ran between 0.20 and 0.45.
The famous retention statistic needs the same treatment. “Increase retention 5% and profits rise 25% to 95%” is quoted everywhere; it comes from Reichheld and Schefter in 2000, restating Reichheld and Sasser’s 1990 HBR analysis of service businesses, whose own summary says profits can rise “by almost 100%” from retaining 5% more customers. Bain’s later published version is narrower still: “In financial services, for example, a 5% increase in customer retention produces more than a 25% increase in profit.” One industry, a lower bound, and no sample or method published. None of it was measured in ecommerce.
So the honest case for a program is behavioral, not arithmetic by assumption. Liu’s longitudinal study of a convenience-store franchise found that light and moderate buyers genuinely increased frequency and basket size after joining, while the program’s heavy buyers “benefited most from the program, but their spending did not grow over time”. That is the whole proposition in one sentence: a program can change the behavior of your middle, and it will pay your top for what they were already doing.
Is a loyalty program right for your store?
It depends on purchase frequency, margin headroom, and whether your competitors already run one. The last of those is the factor most readiness checklists omit, and it can erase the benefit entirely.
The frequency test is straightforward. A program needs a plausible second, third and fourth purchase to reward, so consumables, beauty, pet, coffee and fashion are natural fits, while rare big-ticket purchases are not. For those, warranty, service and referral usually do more than points.
The margin test is arithmetic. Your effective reward rate is the redeemed value divided by the revenue it was earned on, and it has to sit inside your gross margin with room left over. Model it at full redemption, not at expected redemption, so a sudden redemption wave is not a surprise.
The competitive test is the one worth adding. Research across 358 grocery banners in 27 countries found the positive effect of basic direct rewards disappeared in markets where 75% or more of competitors ran a program. Dowling and Uncles made the same argument in 1997: “it is probably a mistake to introduce a frequent-buyer program if you are selling a parity brand in a competitive market.” If everyone in your category has points, points are table stakes rather than an advantage, and the money may be better spent on product or service.
What are the main loyalty program structures?
Points, tiers and cashback or store credit cover most ecommerce programs, with referral, paid and subscription models alongside them. Each works through a different mechanism, and the mechanisms have real research behind them.
Points work partly as a currency people optimize for its own sake. Hsee and colleagues showed this directly: in one study 44% chose a longer errand when points were attached against 17% without, and in another 61% against 30%. The finding that should shape your design is the boundary condition, because the effect disappears when the point-to-value mapping is nearly linear or transparently trivial. That is the real reason earn and redeem rates need designing rather than guessing.
Tiers work through status and through goal proximity. Firm-level evidence favours them: across 322 public companies introducing programs between 2000 and 2015, designs with tiers or earning mechanisms produced significantly larger gains in sales and gross profit than simpler ones. The risk is the mirror image, and it is measurable. In a 329-respondent study, 56% took up a promotional offer when close to a reward against 38% when far from it, so a top tier that reads as unreachable actively suppresses response.
Cashback and store credit are the simplest to understand and the easiest to model, with one accounting caveat covered below. Credit is redeemed against a margin-bearing basket rather than paid out, which is a genuine advantage over a cash-equivalent giveaway. The full comparison is in points versus tiers versus cashback.
How do I design rewards that actually drive behavior?
Make the first reward feel close, and keep the ladder short enough to stay credible. Both of those are experimental findings rather than opinions, and the experiments are worth knowing because they tell you what to manipulate.
The strongest single result is the endowed progress effect. Nunes and Dreze ran a car-wash field experiment with 300 customers in which both groups had to buy exactly eight washes. One group got a ten-space card with two stamps already filled in; the other got an eight-space card with none. Completion was 34% against 19%. The required effort was identical; only the perceived starting position differed. The practical translation is to grant status or points at enrollment rather than starting everyone at zero.
The second is goal-gradient acceleration. Kivetz, Urminsky and Zheng analyzed 949 completed ten-stamp cafe cards, roughly 10,000 purchases, and found interpurchase time shortened by a mean 0.7 days, about a 20% acceleration, as the reward approached. So the distance to the next reward is the live variable, not the generosity of the top one.
Put those together and the design brief is specific: a reachable first reward, visible progress, and a next step that is always close. An earlier version of this page said a reward that is too distant demotivates and one that is too generous erodes profit, which is correct but was presented as our opinion. It is a published finding and it deserves the citation.
How do I avoid giving away margin?
Reward incrementality rather than existing behavior, keep the effective reward rate inside your margin, and remember that unredeemed points are a liability on your balance sheet rather than a saving.
The incrementality point has a measurement behind it. Liu found that the heaviest buyers claimed the most rewards and did not increase their spending, while light and moderate buyers did change. Meyer-Waarden and Benavent found that heavier customers also enroll earlier. So a program left to run on default settings will route most of its reward budget to the customers least likely to change, which is the single largest source of margin leakage in loyalty.
Then there is the accounting, which loyalty advice almost always omits. Under ASC 606 and IFRS 15, points granted with a sale are a material right and therefore a separate performance obligation: part of the transaction price is allocated to the points and deferred until they are redeemed, with breakage recognized in proportion to the pattern of redemptions, or only when further redemption becomes remote. You cannot recognize it up front. So an outstanding points balance is a real liability, not a float, and a program that accrues faster than it redeems is quietly building one.
On point expiry, note that loyalty points sit outside the US federal gift-card rules: Regulation E expressly exempts “a loyalty, award, or promotional gift card” from the five-year minimum. That gives you latitude, and it also means your expiry terms are governed by your own contract and by general unfair-terms law, so write them clearly and do not devalue retroactively. If the real consideration for your program is customer data, California Civil Code 1798.125 applies: material terms disclosed, prior opt-in consent that can be revoked, and no incentive that is “unjust, unreasonable, coercive, or usurious in nature.”
What are the common loyalty mistakes?
Rules nobody can explain, a first reward that feels unreachable, a reward budget flowing to customers who would have bought anyway, launching quietly, and measuring the program against the wrong comparison group.
That last one is the most consequential and the least obvious, so it gets its own treatment below. The others are covered in detail in loyalty program mistakes.
One claim we have removed: that most programs fail for these reasons. We have no prevalence data, and “most” was doing work no source supports. The documented failure modes in the literature are somewhat different and more structural: competitive saturation neutralizing the benefit, launch effects dissipating within months, and rewards concentrating on customers who were not going to change.
How do I know whether the program is working?
Not by comparing members to non-members. That comparison is confounded by who chooses to join, and it will credit the program with behavior it did not cause. Use a matched comparison, a within-customer baseline, or a randomised holdout instead.
The problem is well documented. Meyer-Waarden and Benavent found that “heavier, more frequent store customers enroll in the loyalty program earlier” and that “buying behavior changes only slightly after buyers join the program”. Leenheer and colleagues found that once self-selection is explicitly modeled, the effect on share of wallet is only small and positive, far below the naive member-versus-non-member gap. Your best customers join first, so the gap mostly measures who signed up.
Three workable alternatives. Compare members to non-members matched on pre-enrollment purchase behavior. Compare each member to their own pre-join baseline, which controls for the selection directly. Or, best of all, hold back a randomised share of the enrollment invitation and compare invited to uninvited. The last one is the only design that gives you a clean causal read, and it costs nothing but discipline at launch.
Whichever you use, judge it on incremental gross margin after reward cost, not on sign-ups or on member revenue share. And expect profit to lag sales: firm-level work on 322 companies found gross-profit effects arriving a quarter after the sales effect and being much smaller. A program that looks like it is working in month two may simply be accruing a liability.
We design loyalty around real repeat behavior and real margins, then measure it.
We check whether a program fits your purchase frequency and margins, pick a structure that drives the behavior you want, set attainable rewards funded by incremental repeat purchases, and measure repeat rate, frequency, and margin, so loyalty builds profit instead of giving it away.
It’s the same measurement-first discipline behind our conversion optimization and lifecycle work, where we rebuild the buying path and measure the result .
// we would rather show up in your reporting than in your inbox
Frequently asked questions
Do repeat customers really cost less to sell to?
The evidence does not support it as a general rule. Reinartz and Kumar studied four companies and more than 16,000 customers over four years and found no meaningful cost-to-serve advantage: 6.3 cents against 6.5 cents per sales dollar at the mail-order company, and at one firm loyal customers were more expensive. They also paid 5% to 7% less. What is better supported is that existing customers can be made to buy more often, which is a different claim.
Is a loyalty program worth it for my store?
It depends on three things. Whether your customers can plausibly buy repeatedly, whether your gross margin can carry the reward rate at full redemption, and whether your competitors already run programs. On the last point, research across 358 grocery banners in 27 countries found the benefit of basic direct rewards disappeared where 75% or more of competitors had a program.
Should I use points, tiers or cashback?
Points are the most common and the easiest to launch, and they work partly because people optimize the currency itself, an effect that collapses if the point-to-value mapping is transparently trivial. Tiers have the strongest firm-level evidence, with tiered designs producing significantly larger sales and gross-profit gains across 322 companies, but an unreachable top tier suppresses response. Cashback and store credit are the simplest to model.
How do I make rewards motivating without overspending?
Make the first reward feel close. In a 300-customer car-wash field experiment, 34% completed a card that started with two stamps already filled in against 19% starting from zero, for exactly the same eight required purchases. And keep the next reward near: analysis of 949 cafe cards found purchases accelerating about 20% as a reward approached. Grant progress at enrollment rather than starting everyone at nothing.
How should I measure whether the program works?
Not by comparing members to non-members, because your best customers join first and that gap mostly measures self-selection. Compare members to non-members matched on pre-enrollment behavior, compare each member to their own pre-join baseline, or hold back a randomised share of the enrollment invitation. Judge it on incremental gross margin after reward cost, and expect profit to lag sales by at least a quarter.
Sources
Reinartz & Kumar, “The Mismanagement of Customer Loyalty,” Harvard Business Review, July 2002. Customer databases of four companies, more than 16,000 individual and corporate customers, over four years: a high-tech corporate service provider, a large US mail-order company, a French grocery retailer and a German direct brokerage. Findings that bear directly on the claim that loyal customers are cheaper: the only strong longevity-to-cost correlation suggested loyal customers were more expensive to serve; at the mail-order company communication cost per dollar of sales was 6.3 cents for long-standing against 6.5 cents for newer customers; long-term customers paid 5% to 7% less at the corporate service provider and regular customers 9% less in one mail-order category; and the authors conclude that “a loyal customer … is actually more price sensitive than an occasional one.” Longevity-to-profitability correlations ran 0.20 to 0.45. HBR: The Mismanagement of Customer Loyalty
Reichheld & Sasser, “Zero Defections: Quality Comes to Services,” Harvard Business Review, September 1, 1990. Bain’s own summary of the article states that companies “can boost profits by almost 100% by retaining just 5% more of their customers.” The analysis is of service businesses, the single named example is MBNA, a credit-card issuer, and the public material describes no sample, statistical test or derivation. The widely quoted “25% to 95%” range is not from this article; it appears in Reichheld & Schefter’s 2000 HBR piece as a restatement of the same 1990 services analysis. Bain research on customer retention (Zero Defections)
Fred Reichheld, Bain & Company, “Prescription for cutting costs.” States the narrower version: “In financial services, for example, a 5% increase in customer retention produces more than a 25% increase in profit.” One industry, a lower bound only, with no sample size or period published. Neither this nor the 1990 article measured ecommerce. Bain research on retention economics
Liu (2007), “The Long-Term Impact of Loyalty Programs on Consumer Purchase Behavior and Loyalty,” Journal of Marketing 71(4):19-35. Longitudinal convenience-store franchise data: the program’s heavy buyers “benefited most from the program, but their spending did not grow over time,” while light and moderate buyers genuinely increased purchase frequency and transaction size. This is the measured version of the criticism that programs pay the already-loyal for behavior they already had. Peer-reviewed study (SAGE) on design an ecommerce loyalty program
Nunes & Dreze (2006), “The Endowed Progress Effect,” Journal of Consumer Research 32(4):504-512. Car-wash field experiment, 300 customers, identical eight paid washes required either way: 34% completed a ten-space card given two stamps already filled in, against 19% given an eight-space card with none. Artificial advancement toward a goal raised completion by framing alone. Peer-reviewed study (Oxford) on design an ecommerce loyalty program
Kivetz, Urminsky & Zheng (2006), “The Goal-Gradient Hypothesis and Its Discontents,” Journal of Marketing Research 43(1):39-58. Cafe-card field data: 949 completed ten-stamp cards, roughly 10,000 purchases, with interpurchase time shortening by a mean 0.7 days (t = 2.6, p < .05), about a 20% acceleration, as the reward neared. A companion frequent-flyer study of 329 respondents found 56% accepted a promotional offer in the near-goal condition against 38% in the distant-goal condition (chi-square = 10.7, p < .001). Peer-reviewed study (SAGE) on design an ecommerce loyalty program
Hsee, Yu, Zhang & Zhang (2003), “Medium Maximization,” Journal of Consumer Research 30(1):1-14. People optimize the points rather than the underlying outcome: in Study 2 (213 analyzed) 44% chose the longer errand in the points condition against 17% without points; in Study 4 (105) it was 61% against 30%. Critically, the effect disappears when the point-to-value mapping is nearly linear or transparently trivial, which is the real reason earn and redeem rates have to be designed rather than guessed. Peer-reviewed study (Oxford) on design an ecommerce loyalty program
Meyer-Waarden & Benavent (2009), Journal of the Academy of Marketing Science 37(3):345-358. “Heavier, more frequent store customers enroll in the loyalty program earlier,” and “buying behavior changes only slightly after buyers join the program,” with small loyalty gains eroding six to nine months after joining. This self-selection is why comparing members to non-members overstates program effect. Peer-reviewed study (Springer) on design an ecommerce loyalty program
Leenheer, van Heerde, Bijmolt & Smidts (2007), International Journal of Research in Marketing 24(1):31-47. Once self-selection into the program is explicitly modeled, the effect of loyalty-program membership on share of wallet is small and positive, far below the naive gap between members and non-members. The methodological point that follows: a member-versus-non-member comparison is not a measurement of the program. Peer-reviewed study (ScienceDirect) on design an ecommerce loyalty program
Revenue recognition for loyalty points: ASC 606 and IFRS 15. Points granted with a sale are a material right and therefore a separate performance obligation: part of the transaction price is allocated to the points and deferred until redemption, with breakage recognized in proportion to the pattern of redemptions (ASC 606-10-55-46 to 55-48) or only once further redemption becomes remote. Under 55-49, amounts state escheat law requires to be remitted are a liability rather than revenue. IFRS 15 allocates on the same standalone-selling-price basis (IFRS 15.77 to 80). RevenueHub on gift card breakage
Regulation E, 12 CFR §1005.20(b) (exemptions). The federal gift-card rules, including the five-year minimum before funds may expire, expressly exclude “a loyalty, award, or promotional gift card.” So loyalty points are not covered by that five-year floor, and point expiry is governed instead by your own terms and by general unfair-terms and consumer-protection law. US Regulation E gift card rules
California Civil Code §1798.125 (CCPA as amended by the CPRA), financial incentives. A business may offer a financial incentive, including a loyalty program, for the collection, sale or retention of personal information, but must notify the consumer of the material terms, obtain prior opt-in consent that may be revoked at any time, and may not use an incentive that is “unjust, unreasonable, coercive, or usurious in nature.” Relevant to any program whose real consideration is customer data. California CCPA non-discrimination (1798.125)
Shopify, “Ecommerce Loyalty Program: 7 Ways to Build One,” published March 4, 2024. Vendor marketing content, cited only for what it contains: the seven program types (cash-back, points, referral, tiered, paid, subscription and hybrid) and their definitions. It carries no cost-to-serve data, so we no longer cite it for the claim that repeat customers cost less to sell to, which it does not support. Shopify guide to loyalty programs