Ecommerce Loyalty Program ROI: Why Your Program Is Not Building the Loyalty You Think It Is

A skincare brand pulls its quarterly numbers and finds that loyalty members spend 41 percent more per year than everyone else. The program gets renewed on the spot. Nobody in the room asks the one question that would have changed the decision: were those customers already the highest spenders before they ever signed up?

That single unasked question is why ecommerce loyalty program ROI is one of the most confidently reported and least reliably measured numbers in retention marketing. The number on the slide is real. What it proves is not.

This is not an argument that loyalty programs do not work. Some do, and the research shows exactly where and for whom. It is an argument that the way almost every brand measures its program guarantees a flattering answer regardless of the truth. This guide breaks down what the independent evidence actually says, why vendor-published data tells a very different story, what your points balance is really costing you, and the specific testing method that produces a number you can defend in a board meeting.

What Ecommerce Loyalty Program ROI Actually Measures (And What Most Brands Measure Instead)

Ecommerce loyalty program ROI is the incremental margin your program generates minus its total cost, divided by that cost. The word doing all the work in that sentence is incremental: revenue that would not have existed without the program. Everything else is revenue you already had, wearing a badge.

Most brands calculate something entirely different. They total up what loyalty members spent, subtract what the program cost to run, and call the difference a return. That method counts every repeat purchase from every member as a program win, including purchases from customers who would have bought anyway, at full price, with no points involved.

Why “Members Spend More” Is Not Evidence Your Program Works

Think about who signs up. It is the customer who has already bought three times and already intends to buy again. Enrollment signals existing loyalty. It does not create it.

The comparison brands run, member spend against non-member spend, therefore measures two things at once and cannot separate them. It captures whatever the program genuinely caused, plus the pre-existing difference between people who opt in and people who do not. Statisticians call this self-selection bias. In loyalty measurement it is not a small distortion at the edges. It is the majority of the number.

Here is the practical version. Suppose your members generate an average annual value of 340 dollars and your non-members generate 240 dollars. The intuitive read is that the program is worth 100 dollars per member per year. The correct read is that you have no idea what the program is worth, because you have not compared members to a group of customers who look identical to them and simply were not offered the program.

The takeaway is uncomfortable but clarifying. If your program reporting relies on any metric in the left-hand column, you do not currently know your ecommerce loyalty program ROI. You know your member revenue, which is a different and far less useful number.

The Self-Selection Problem: Your Best Customers Join, Your Program Does Not Create Them

The most rigorous test of this question comes from academic marketing science rather than the industry press, and the finding is blunt.

What Happens When Researchers Correct for Self-Selection

In a study published in the International Journal of Research in Marketing (Leenheer, van Heerde, Bijmolt and Smidts, 2007), researchers analyzed household panel data covering every major grocery loyalty program in a national market. They ran the naive comparison first, the same one brands run internally, and then re-ran the analysis with a statistical correction for the fact that customers choose whether to join.

The corrected effect on share of wallet was positive and significant, but seven times smaller than the naive model suggested. Seven times. The genuine lift came out at roughly four percentage points of share of wallet, against a raw figure that implied something far more dramatic.

Read that as a translation exercise. If your internal dashboard says the program is responsible for a 30 percent spend lift, the defensible estimate after correcting for who joins is closer to a low single-digit gain. That is not nothing. It is also not what the budget was approved on.

The Double Jeopardy Law and Why Loyalty Follows Penetration

There is a second, older problem underneath the first. Work from the Ehrenberg-Bass Institute established that repeat-purchase loyalty is highly predictable from a brand’s market penetration alone. Bigger brands get more buyers and slightly more loyal ones, and smaller brands get fewer of both. This is the Double Jeopardy Law, and it means a large share of the loyalty you observe was never available as a lever in the first place.

Applied directly to programs, Sharp and Sharp (International Journal of Research in Marketing, 1997) benchmarked six loyalty-program brands against the repeat-buying levels statistical models predicted they should achieve anyway. Only two showed any excess loyalty at all, and those deviations turned up among non-members too, which made them hard to attribute cleanly to the program.

The pattern repeats in customer profitability research. Reinartz and Kumar (Harvard Business Review, 2002) studied 16,000 customers across four companies’ databases and found no support for the standard claims that loyal customers cost less to serve, pay higher prices, or reliably market by word of mouth. Their conclusion was that not all loyal customers are profitable and not all profitable customers are loyal, which is precisely the distinction a points balance cannot see. If your retention program is built on the assumption that loyalty and profitability move together automatically, it is worth revisiting how the broader retention system is supposed to fit together before adding another mechanic on top.

Do Loyalty Programs Increase Retention? What the Independent Evidence Says

Yes, but conditionally, and rarely for the customers most programs are designed around. This is where an honest reading of the research complicates the cynical version of the story, and it is worth stating plainly rather than skipping past.

Where Programs Genuinely Create Incremental Value

The clearest positive finding comes from Liu (Journal of Marketing, 2007), who tracked purchase behavior over time and found no measurable effect on heavy buyers, the customers already deeply committed to the brand, alongside sustained behavioral improvement among light and moderate buyers. The program worked. It just worked on the segment most brands under-invest in, while the rewards flowed disproportionately to the segment where they changed nothing.

Mechanism research supports this. Taylor and Neslin (Journal of Retailing, 2005) separated two distinct effects in a retail frequency program: a six percent storewide sales lift while customers accumulated toward a reward, and a 17.5 percent weekly spend lift among customers after they redeemed one. Both are real. Both depend entirely on customers actually reaching redemption, and only about a fifth of them did.

The strongest tier of evidence agrees. Belli and colleagues (Journal of the Academy of Marketing Science, 2022) synthesized 429 effect sizes across three decades and concluded that programs do enhance loyalty overall, with behavioral loyalty responding more consistently than attitudinal loyalty.

Where the Evidence Says They Do Not

The catch is that “programs work on average” and “your program is working” are unrelated claims. Averages in that meta-analysis include well-designed programs measured properly. They tell you nothing about whether a generic points-per-dollar scheme, launched because a competitor had one, is producing anything beyond a rolling discount.

Competitive context matters more than most brands assume. Bombaij and Dekimpe (International Journal of Research in Marketing, 2020) examined 358 grocery banners across 27 countries and found that programs lifted sales productivity in some markets and produced no effect whatsoever in markets where more than three quarters of competitors also ran one. Program saturation neutralized the advantage entirely, and it mattered more than any individual design decision.

So the honest answer to whether loyalty programs increase retention is this: they can, in specific segments, in markets where they are not table stakes, when the reward mechanics genuinely change behavior rather than rewarding behavior that was already happening. Every one of those conditions is testable. Almost none of them get tested.

The Vendor Data Problem: Why Every Number You Have Read Is Positive

Research this topic and you hit a wall of encouraging statistics. Programs return 4.8x. Programs return 5.3x. Those figures are not fabricated. They just come from a specific place.

Antavo’s Global Customer Loyalty Report (2026), a survey run by a loyalty software company, found that 92.7 percent of program owners who measure ROI report a positive return, averaging 5.3x. Compare that against McKinsey’s independent observation that roughly two thirds of established programs fail to deliver value. Both statements can be technically accurate at the same time, because they are not measuring the same thing.

The vendor figure is self-reported by marketers using the member-versus-non-member method described earlier. The consultancy figure comes from outside assessment of whether programs created value net of their cost. One measures perception built on a flawed comparison. The other measures outcome.

None of this makes vendor data useless. It makes it directional rather than decisive, and it means a benchmark published by a company that sells loyalty software should never be the basis for renewing a loyalty budget. The point is not that vendors are dishonest. It is that nobody publishes research proving their own product is optional, and the cost of that blind spot compounds quietly across every quarter the program runs unexamined.

Points Are a Liability, Not a Perk

There is a second cost to loyalty programs that marketing dashboards almost never surface, and finance teams almost never forget.

Under ASC 606 in the United States and IFRS 15 internationally, points issued to a customer represent a performance obligation you have not yet fulfilled. The portion of the sale attributable to those points cannot be recognized as revenue at the time of purchase. It sits on the balance sheet as deferred revenue until the customer redeems, or until the points expire and you recognize breakage.

The scale becomes obvious at a large enough company. Starbucks disclosed roughly 1.75 billion dollars in stored-value and loyalty program deferred revenue in its fiscal 2025 filings with the SEC. That is not a marketing metric. It is a line item auditors examine.

Why Redemption Spikes Are Not Wins

Marketing teams tend to celebrate redemption activity as proof of engagement. Finance sees the same event as a liability converting into a margin cost.

Both readings are valid, but only one of them belongs in an ROI calculation, and it is the finance one. A redemption is a discount you promised earlier and are paying for now. When redemption clusters around purchases the customer was going to make regardless, and the research suggests a substantial share does, you have converted a balance-sheet liability into a straight margin reduction with no behavior change attached.

The Number That Belongs in Your ROI Calculation

Reward costs typically consume somewhere in the range of three to seven percent of revenue for programs at scale, before you account for platform fees, integration work, and the internal time spent managing tiers and campaigns. Breakage, the share of points that expire unredeemed, commonly runs between 15 and 33 percent in retail, which quietly softens the reported cost while inflating the liability you carry in the meantime.

A complete cost base for your ecommerce loyalty program ROI therefore includes four things: the margin cost of redeemed rewards, the platform and integration spend, the internal labor to operate the program, and the carried liability of outstanding points. Most brands count the second item and forget the rest.

Who Your Loyalty Program Is Actually Attracting

Enrollment numbers create a comforting illusion of momentum, and the underlying membership data does not support it.

The Bond Brand Loyalty Report (2025) found consumers hold an average of 17.4 memberships while actively using only 8.8 of them, with active membership declining year over year. Roughly half of every program’s roster is inert. Those dormant accounts still sit in your member count, still inflate your member-versus-non-member comparison, and still cost you nothing to keep, which is exactly why nobody removes them.

Beneath the dormancy sits a sharper problem. Lewis (Journal of Marketing Research, 2006) found customers acquired through a deep promotional offer were worth roughly half as much in long-term asset value as those acquired without one.

A more recent study makes the same point inside a live program. Analyzing 210,657 loyalty members, Nishio and Hoshino (Journal of Retailing and Consumer Services, 2024) found that customers who received a birthday reward showed materially lower lifetime value than a control group, around 72 dollars against 125 dollars. The incentive attracted response from more impulsive, lower-value buyers rather than deepening relationships with the valuable ones.

This is the same dynamic that makes discount-led growth look healthier than it is, and it is why lifetime value has to be the reporting unit rather than campaign revenue. It also explains why loyalty enrollment so rarely moves the metric that actually funds a business, which is the compounding value of customers you keep at full price.

One more distinction worth drawing carefully: a dormant member is not the same thing as a churning customer. Someone can stop engaging with your points balance while continuing to buy from you happily, and someone can hold a healthy points balance while quietly disengaging from the brand entirely. Program activity is a poor proxy for relationship health, which is why the early warning signals of silent churn need to be tracked independently of loyalty status.

How to Measure Ecommerce Loyalty Program ROI Correctly

Correct measurement requires comparing your program against a version of reality where it does not exist. That sounds abstract. In practice it is a holdout group, and it is entirely achievable for a mid-sized ecommerce brand.

The 90-Day Holdout Test

If you are launching or relaunching a program, this is the cleanest method available.

Randomly split new customers into two groups. Offer enrollment to half. Withhold the offer entirely from the other half, who receive your normal marketing with no program mechanics at all. Run the split for 90 days minimum, longer if your purchase cycle is slow, and then compare the two groups on repeat purchase rate, average order value, total revenue per customer, and gross margin per customer.

The difference between those groups is your incremental lift. Not the difference between joiners and non-joiners, which is contaminated by choice, but the difference between two randomly assigned populations. That number, minus your full cost base, is your real ecommerce loyalty program ROI.

What to Do If Your Program Is Already Live

Most brands reading this cannot run a clean launch experiment because the program has existed for years. There are two workable alternatives.

The first is matched-cohort analysis. Build a comparison group of non-members who resemble your members on the variables that predict spend: acquisition channel, first-order value, product category, tenure, and pre-enrollment purchase frequency. Compare only against that matched group. It is imperfect, since you cannot match on intent, but it removes the crudest layer of selection bias.

The second is a pre-post behavioral analysis. Compare each member’s purchase behavior in the 180 days before enrollment against the 180 days after. That asks a far better question: did this customer change, or carry on exactly as before while collecting points for it?

The Threshold That Decides Whether to Scale or Restructure

Once you have an incremental number, the decision becomes straightforward.

If incremental margin sits comfortably above your total program cost, including reward margin, platform fees, labor, and carried liability, the program is creating value and deserves more investment. If incremental margin falls within the noise range of your test, the program is not creating loyalty. It is subsidizing behavior you already had, and the budget should be redirected toward mechanics with a measurable effect.

Worth noting that EY’s 2025 loyalty research found a substantial share of loyalty leaders struggle to quantify their program’s overall impact at all. Simply having a defensible incremental number puts you ahead of most of the market, regardless of what that number turns out to be. Designing that test correctly is part of the analytics work that should sit underneath any retention program.

What Works Instead of Points

If the evidence against generic points programs is strong, the evidence for specific alternatives is worth acting on.

Non-monetary rewards outlast monetary ones. Melnyk and Bijmolt (European Journal of Marketing, 2015) studied what happened when programs were introduced and later terminated, and found that monetary savings showed no significant effect on loyalty at either point, while non-monetary elements such as recognition, member-only access, and events both built loyalty and sustained it after removal. A discount creates a habit that dies with the discount. Status does not.

Paid membership consistently outperforms free enrollment. McKinsey (2020) found members of paid loyalty programs were 60 percent more likely to increase spending with a brand after joining, against 30 percent for free programs. The mechanism is not complicated. Paying for membership requires a deliberate decision, which creates commitment, and it forces the brand to deliver value continuously rather than accumulating an obligation it hopes will expire.

Access now outperforms discounts on preference. Bond’s 2025 research found exclusive experiences overtook financial rewards as the top driver consumers cited, a first in a decade.

Finally, reconsider who your tiers are built for. Most programs concentrate their richest rewards on the highest spenders, precisely the group Liu’s research found least responsive. Tiering on behavior change rather than raw spend, rewarding a second purchase or a category expansion rather than a cumulative dollar threshold, targets the light and moderate buyers where the incremental effect actually lives.

Loyalty Program ROI Self-Audit Checklist

Run your current program against these nine checks. Three or more failures means your reported ROI is almost certainly overstated.

1. Your ROI calculation uses incremental margin against a control group, not total member revenue.
2. You have run a holdout test, a matched-cohort analysis, or a pre-post behavioral comparison within the last 18 months.
3. Dormant members are excluded from your active membership count and your performance reporting.
4. Your cost base includes reward margin, platform fees, internal labor, and outstanding points liability.
5. You track lifetime value separately for members acquired through an incentive versus members who joined organically.
6. Your tier thresholds reward behavior change, such as a second purchase or category expansion, rather than cumulative spend alone.
7. At least one meaningful reward in your program carries no margin cost, such as early access or member-only launches.
8. You know your redemption rate and how it compares to the 20 to 30 percent range typical in ecommerce.
9. Your finance team and your marketing team can both explain the program’s cost using the same number.

Frequently Asked Questions

How do you calculate ecommerce loyalty program ROI?

Divide incremental margin generated by the program by the total cost of running it. Incremental margin means profit that would not have existed without the program, which requires comparing enrolled customers against a control group of similar customers who were not offered the program. The total cost includes reward margin, platform fees, internal labor, and the outstanding liability of unredeemed points. Comparing member revenue against non-member revenue produces a number, but it is not ROI, because it counts purchases that would have happened anyway.

Do loyalty programs increase retention?

They can, but the effect is far smaller than raw comparisons suggest and it is concentrated in specific segments. Research published in the Journal of Marketing in 2007 found no measurable effect on heavy buyers who were already loyal, alongside sustained improvement among light and moderate buyers. A separate 2007 study correcting for self-selection found the true share-of-wallet effect was seven times smaller than uncorrected analysis implied. Programs also lose their effect in markets where most competitors run one, because the advantage becomes table stakes rather than a differentiator.

What is a good loyalty program redemption rate?

Ecommerce programs typically see redemption rates in the 20 to 30 percent range, below the roughly 50 percent average across all loyalty categories. A very low rate signals that rewards are too hard to reach, so you carry liability without changing behavior. A very high rate is not automatically good either.

Why do most loyalty programs fail?

The most common failure is a design that rewards existing behavior rather than changing it, combined with measurement that cannot detect the difference. McKinsey has observed that around two thirds of established programs fail to deliver value, with many eroding it. Programs also fail when they are launched to match a competitor, since research across 27 markets found the sales benefit disappears entirely once most competitors in a category offer a program. A third common cause is cost blindness, where the margin cost of rewards and the carried points liability never appear in the same report as the revenue.

Are unredeemed loyalty points a liability?

Yes. Under ASC 606 in the United States and IFRS 15 internationally, points issued represent an unfulfilled performance obligation, so the portion of a sale attributable to them is deferred rather than recognized as revenue at the point of purchase. It remains a liability on the balance sheet until the customer redeems or the points expire, at which point the business recognizes breakage. Retail breakage commonly runs between 15 and 33 percent, and large programs can carry liabilities in the hundreds of millions or billions of dollars.

Measure It Properly Before You Judge It

The uncomfortable conclusion here is not that loyalty programs are worthless. It is that most brands have never generated the evidence required to know either way, and the metric they trust most is the one specifically designed to hide the answer.

Three things follow from that. Run a real incrementality test, whether that is a holdout at launch or a matched cohort on an existing program. Put the full cost, including carried points liability, in the same report as the revenue. Then redirect reward investment toward the light and moderate buyers where the research says behavior actually changes, using access and recognition rather than another stored discount.

Do that and you will end up with a number you can defend. It may be smaller than the one on your current slide. It will also be the first one that tells you something true about your ecommerce loyalty program ROI, and a smaller honest number is a far better foundation for a budget decision than a large number nobody has tested.

Ready to Turn Your Customers Into Loyal Buyers?

OrangeFox helps e-commerce brands add 15 to 25 percent to their revenue through data-driven retention marketing. Let us show you exactly where your brand is losing revenue, and how to fix it.

Book Your Free Retention Audit →