Open your wallet, real or digital, and count the loyalty cards. Now ask yourself how many of those brands you'd happily abandon tomorrow for a better deal. That gap, between the card you carry and the loyalty you actually feel, is the whole subject of this explainer, and it's where a lot of marketing budget quietly disappears.

The quick version

  • Engagement is how actively a customer interacts with you; retention is whether they keep buying; a loyalty program is a structured incentive meant to nudge both. They're three different things and people routinely confuse them.
  • Loyalty has two halves: behavioural (they keep buying) and attitudinal (they actually prefer you). A program can buy the first while doing nothing for the second, which is fragile loyalty you're renting, not building.
  • The evidence is humbling: large studies find loyalty programs produce only a small "excess loyalty" effect, and mostly reward heavy buyers who were already loyal.
  • Retention pays off hard when it's real, but the cause is usually a good product, low friction and fair treatment, not the points. Fix those first; bolt on a program only if it earns its keep.

The idea in depth: loyalty is two things, and you can buy only one of them

Start with the cleanest academic frame, because it stops most of the confusion before it starts. In 1994, Alan Dick and Kunal Basu published Customer Loyalty: Toward an Integrated Conceptual Framework in the Journal of the Academy of Marketing Science, and their core move was to split loyalty along two axes: a customer's relative attitude (how strongly they prefer you over the alternatives) and their repeat patronage (whether they actually keep buying). Cross those two and you get four states, not one.

The dangerous quadrant is what they called spurious loyalty: high repeat-buying, low attitude. The customer keeps coming back, but out of habit, convenience, a contract, or a points balance they don't want to forfeit, not because they'd choose you on a level field. It looks like loyalty in the sales data and evaporates the moment a rival removes the friction. The opposite trap is latent loyalty: they love you but rarely buy, usually because something in the way is blocking them. The practical move: before you spend a cent on a program, work out which quadrant your "loyal" customers are actually in. If your repeat business is spurious loyalty held up by switching costs, a points scheme just deepens a dependency that isn't preference, and you'll feel it the day a competitor makes leaving easy.

flowchart TD
  Start("How loyal is this customer, really?") --> Att("Do they actually prefer us? (attitude)")
  Att -->|High| A2("Do they keep buying? (behaviour)")
  Att -->|Low| B2("Do they keep buying? (behaviour)")
  A2 -->|High| True(["True loyalty: preference + repeat"])
  A2 -->|Low| Latent(["Latent loyalty: they like us but something blocks the buy"])
  B2 -->|High| Spurious(["Spurious loyalty: habit or lock-in, no real preference"])
  B2 -->|Low| None(["No loyalty: up for grabs"])
Dick & Basu's two axes: repeat-buying alone can hide a customer who doesn't actually prefer you. Leaders Loop

Now the inconvenient evidence about programs themselves. The first large-scale empirical test of whether loyalty programs change buying behaviour came from Byron Sharp and Anne Sharp, Loyalty Programs and Their Impact on Repeat-Purchase Loyalty Patterns (International Journal of Research in Marketing, 1997). Using panel data and Dirichlet modelling to control for the fact that loyal customers self-select into programs, they found only a small "excess loyalty" effect, the program nudged repeat-buying a little above what the market norm predicted, but nowhere near the transformation the business case usually assumes. Later work from the Ehrenberg-Bass Institute, summarised in Sharp's How Brands Grow (2010), reinforced the deflating point: brands grow mainly by reaching more buyers, not by squeezing more loyalty from existing ones, and loyalty programs disproportionately reward the heavy buyers who were going to keep buying regardless. Here's what that should change for you: judge a program against the right baseline. The question isn't "did members buy more than non-members" (they always do, loyal people join loyalty schemes). It's "did members buy more than they would have without the program", and you can only answer that with a holdout group you deliberately exclude.

"Loyalty programs produce a small amount of excess loyalty.", Sharp & Sharp's finding, replicated since

So why do retention and loyalty get talked about as if they print money? Because when loyalty is real, the economics are genuinely strong, people just attribute the gains to the wrong cause. Fred Reichheld's work at Bain & Company popularised the figure that improving customer retention by 5% can lift profits by 25% to 95%, depending on the industry, because a retained customer costs less to serve, buys more over time, and refers others (The Loyalty Effect, 1996). Reichheld later distilled the whole idea into a single diagnostic, the Net Promoter Score, in his 2003 Harvard Business Review article The One Number You Need to Grow, arguing that "would you recommend us?" predicted growth better than fat satisfaction surveys. The retention payoff is real. But notice what drives it: a customer recommends you because the product and the experience earned it, not because they're accumulating points. So treat retention as an outcome to diagnose, not a lever to pull directly. Measure who'd recommend you and why, fix the reasons people leave, and let the loyalty follow. A program is the last 10% of that work, not the first.

The honest limitation

None of this means loyalty programs never work, it means they work in narrower conditions than the brochure suggests, and the evidence is contested at the edges. Sharp and Sharp's data came largely from frequently bought consumer goods; in categories with high switching costs and rich data, airlines, hotels, some financial services, a well-run program can shift share and fund itself on the first-party data alone. There's also a live commercial counter-argument that modern programs earn their return by pulling light buyers into more frequent purchase rather than over-rewarding heavy ones. The honest reading is that "loyalty programs don't work" is as wrong as "every brand needs one." Know which case you're in: if your category is high-frequency and low-friction, a program is mostly a discount to people who'd stay anyway; if it's high-consideration and the first-party data is valuable, it can be a genuine asset. Either way, the program is never the strategy, it's an instrument you tune against what actually drives repeat business in your market.

A worked example

Take a mid-sized online homewares retailer, a composite, and the figures below are illustrative. Repeat purchase is sliding, so leadership reaches for the default fix: a points program, 1 point per dollar, a $10 reward at 200 points. Six months in, members are outspending non-members and the dashboard looks like a win. It isn't. The members are simply the people who already loved the brand and signed up first; with no holdout group, there's no way to know the program changed anything. Worse, the rewards flow to the heaviest buyers, the customers least at risk of leaving.

Run the toolkit instead. Diagnose the quadrant first: survey lapsed customers and discover the real churn driver isn't a lack of rewards, it's a clumsy returns process and slow delivery. That's latent loyalty being blocked, not weak attitude. Fix the friction: free, no-questions returns and faster dispatch, the things customer-needs work would surface as the actual job to be done. Measure the right thing: add a recommend-question (NPS-style) and a deliberate holdout group so any future program can be judged against a real counterfactual. Then, and only then, design the program, and aim it at the segment that matters: occasional buyers nudged toward a second and third purchase, where the growth actually lives, rather than discounts sprayed at people already buying weekly.

flowchart LR
  subgraph Default["The 'just launch points' instinct"]
    A("Points scheme for everyone") --> B("Members outspend non-members")
    B --> C(["Looks like a win; rewards the already-loyal; churn unchanged"])
  end
  subgraph Toolkit["The toolkit move"]
    D("Diagnose: why do people actually leave?") --> H("Fix friction: returns, delivery")
    E("Measure: recommend-score + holdout group") --> H
    H --> I("Target light buyers' 2nd & 3rd purchase")
    I --> J(["Retention rises from real causes; any program is now measurable"])
  end
Illustrative: points-for-everyone rewards existing loyalty; diagnosing churn and targeting light buyers grows it. Leaders Loop

The illustrative result: same budget, very different outcome. The default version buys a flattering dashboard and changes nothing; the toolkit version lifts retention because it fixed why people were leaving, and can prove any program's effect, because someone built a control group. This is where the topic connects outward, deciding which customers to win back and grow is the work of segmentation and targeting, and turning that into price, channel and experience is the marketing mix doing its job.

Frequently asked questions

What's the difference between engagement, retention and loyalty?

Engagement is activity, opening emails, using the app, reading the content. Retention is continued buying over time. Loyalty is the underlying preference that makes retention durable rather than accidental. They correlate but don't move together: a customer can be highly engaged and still churn, or quietly retained for years without ever feeling loyal. Track all three, and don't let a healthy engagement number reassure you about a retention problem.

Do loyalty programs actually increase loyalty?

Modestly, at best, on the current evidence. Sharp and Sharp's 1997 study and the Ehrenberg-Bass research that followed found only a small "excess loyalty" effect, and most of the reward flows to heavy buyers who were already loyal. Programs can still be worth running, for the first-party data, for a share shift in high-switching-cost categories, or to pull light buyers up the frequency curve, but expecting one to manufacture preference where the product hasn't earned it is the common, expensive mistake.

Is it really cheaper to retain a customer than acquire one?

Usually, yes, retained customers cost less to serve, buy more over time and refer others, which is why Reichheld's work links small retention gains to large profit gains. But "cheaper to keep" doesn't mean "keep at any cost." Pouring discounts into customers who'd have stayed regardless destroys margin, and some customers are unprofitable to retain. The skill is keeping the right customers, not all of them.

What's the single most useful loyalty metric?

There isn't a magic one, but a recommend-style question (NPS) plus an actual repeat-purchase rate is a strong, honest pair: one captures attitude, the other behaviour, the two halves of Dick and Basu's framework. NPS has real critics (it's a blunt instrument and easily gamed), so treat it as a conversation-starter that tells you who to ask "why," not a number to manage to.

We're small with no budget, where do we start?

Not with a program. Start by asking lapsed customers why they left and recent buyers why they'd recommend you, that surfaces the real retention levers for free. Then remove the biggest source of friction in buying or returning. Those two moves outperform a points scheme in most small businesses, because they fix causes instead of renting behaviour.

Related in the Toolkit

Where to go next