What a loyalty program is
A loyalty program is a mechanic under which a customer receives a reward for repeat purchases and for identifying themselves at payment. The reward can be material (points, cashback, a discount) or status-based (priority support, early access to sales, free delivery).
A program always has two goals, and it helps to keep them apart:
- Economic — increase purchase frequency, order value and customer lifespan.
- Infrastructural — obtain shopper identification offline and connect offline receipts with online behaviour.
The second goal often turns out to be the more valuable one, especially for chains with a large offline share: without a loyalty card an offline purchase stays anonymous, and building a unified customer profile becomes impossible.
Program types and their economics
| Type | How it works | What it gives | The main risk |
|---|---|---|---|
| Points-based | N% of the order returns as points, points pay for later purchases | A simple, legible mechanic with broad reach | Turns into a permanent discount for people who were buying anyway |
| Tiered (statuses) | Status depends on spend over a period, privileges grow with it | Motivates customers to reach the next tier | Top tiers are expensive to serve, bottom tiers motivate nobody |
| Paid subscription | A fixed fee for delivery, discounts and service | Prepayment and high frequency among subscribers | Strong self-selection: the most frequent shoppers buy the subscription |
| Cashback | Part of the amount returned as money or balance | Transparent value for the customer | Closest to a plain discount, weak effect on habit |
| Coalition | Points earned and spent across partners | Extra traffic from partners | Loyalty to the coalition rather than to the retailer |
How to count the cost. An awarded point is a deferred discount, and that is how it must be treated in the economics:
Program cost (in percentage points of margin) =
accrual rate x point redemption rate
+ variable platform and communication costs
/ revenue from member orders
An illustration of the logic: if 5% of the order value is awarded and 70% of those points are redeemed, the effective discount on member orders is around 3.5 percentage points. At a 25% margin, the program eats roughly a seventh of the gross profit on that flow. The question that follows is simple: does the program generate a revenue lift that exceeds that figure?
Unredeemed points look like savings, but they are a liability towards customers and can be presented for redemption at any time. A financial model built on the assumption that half the points will never be spent breaks at exactly the moment the program becomes popular.
Program metrics
| Metric | How it is calculated | What it shows |
|---|---|---|
| Share of identified purchases | Revenue from identified orders / total revenue | Reach and quality of identification |
| Active members | Members with a purchase in N months / all members | The real size of the program, not the size of the card base |
| Purchase frequency | Orders per customer per period | The main channel of program influence |
| Repeat purchase rate | Customers with 2+ orders / all customers | Customer return |
| Member versus non-member lifetime value | Average revenue over the lifespan by group | An indicator, not proof of effect (see below) |
| Point redemption rate | Redeemed / awarded | The actual cost of the program |
| Incremental effect | The difference against a holdout group | The only metric that answers whether it works |
Why member metrics almost always lie
The standard presentation goes like this: “program members have a 30% higher average order value, buy twice as often and show triple the lifetime value — the program works.” Almost always that conclusion is incorrect.
The reason is selection bias. The people who join are not a random sample of customers but the ones who intended to buy regularly anyway. Someone who walked into the store once for a gift will not fill in a form and install an app. A regular shopper will. High activity is therefore the cause of joining, not its consequence.
The second source of distortion is defining a member through activity. If a member is defined as someone who used the card, then only people who buy end up in the group, while everyone else — including those who left — ends up in the control. The comparison loses all meaning.
The third is discount cannibalisation. Some points are redeemed against purchases that would have happened without the program. In the report this appears as revenue stimulated by the program, when in fact it is a discount granted retrospectively.
There is only one correct approach: an experiment with a control group. From the customers eligible for the program, a holdout is drawn at random and receives neither the invitation nor the points. After three to six months, revenue per customer is compared across the two groups. The difference, less the cost of the points awarded, is the incremental effect.
If a holdout could not be set up at launch, a weaker but still sensible alternative works: experiment not with participation itself but with the parameters — the accrual rate, the tier thresholds, the expiry period for points. Such tests run on a live program and reveal how sensitive the economics are.
Where the line with a personalization platform runs
A loyalty program lives in a separate class of system — a loyalty platform or the corresponding module of the CRM and till circuit. That is where point balances, accrual and redemption rules, statuses, expiry dates and integration with fiscal equipment and billing are managed. It is an accounting system with monetary obligations, and the requirements match.
A personalization platform solves a different problem: what to show a specific person on the site, in the app and in product selections. It does not award points and does not manage statuses.
There is one overlap, and it is a useful one: program data is a strong signal for segmentation. Member status, point balance, time to expiry, recency of the last card-identified purchase — all of these are profile attributes that improve the quality of segments and the precision of RFM analysis. The exchange runs one way: the program supplies the attributes, personalization uses them to select content and recommendations.
Common mistakes
- Launching without an economic model. The accrual rate is chosen to match a competitor, with no calculation of the margin impact and no scenario for a high redemption rate.
- No holdout group. A year after launch there is nothing left to prove the effect with, and closing the program is politically impossible.
- Judging by the number of cards issued. A headline of two million members means nothing without the share active in the last 6-12 months.
- The same reward for everyone. A customer who buys weekly and one who buys yearly get an identical mechanic, although they respond to different incentives.
- A program instead of range and service. Points do not compensate for out-of-stock items, slow delivery and a poor returns policy — they merely make retention more expensive. A program reinforces working retention, it does not replace it.
- Ignoring churn inside the program. A member who stopped buying stays in the base and keeps flattering the reach statistics.