Loyalty programme vs discount for repeat customers: who would have bought anyway?

A loyalty programme rewards repeat purchases with points, tiers or benefits. The usual proof is a table: members spend more and order more often than non-members.

That table has a flaw. Your best customers join first. People who already like your products, buy often and expect to keep buying are the ones who sign up. So members look better before the programme has done anything. This is the selection effect.

Example (illustrative): Say 40% of non-members reorder within a year and 60% of members do. It is tempting to credit the programme with 20 points. But if those members would have reordered at 55% without it, the programme added 5 points, and it may still cost more than that is worth.

Our view: until a programme has been compared with a similar group that did not get it, "members buy more" is a description of who joined, not a result.

The same logic applies to discounts. A customer who was about to reorder will happily use a code. You count the order as a win and give away margin you did not need to give. That is the same question that incrementality asks of ads: what was added over what would have happened anyway?

Vendor case studies are the easiest evidence to find and the weakest. A loyalty platform's case-study pages are written by the company that sells the platform, each about a brand it chose to feature¹. Each story covers one brand, with no comparison group. Useful for ideas, not for proof.

Independent work on this is hard to find for online shops. An academic paper titled "Loyalty programs and their impact on repeat-purchase loyalty patterns" asks exactly what this page asks². We have not found independent research that answers it for ecommerce brands. Do not read any single study across to every online shop. What the question itself makes clear: "members buy more" is not the same as "the programme made them buy more".

For discounts, a university summary of a Journal of Retailing study looks at whether deep promotional discounts work, using transactions from a US grocery chain's physical stores³. That is physical grocery retail, not ecommerce, so use it as a prompt to test rather than a verdict.

The honest reading: we know of no independent ecommerce evidence for either. That is a reason to measure your own.

Programmes come in a few shapes. Points schemes give a fixed reward per spend. Tiers give extra benefits at higher spend. Some add paid memberships or early access. Each has costs:

• Reward cost. Points are a discount you pay later. Redeemed points reduce the margin on the order they are spent on.
• Benefit cost. Free shipping and gifts for top tiers cost real money on every order.
• Running cost. Software, set-up and the time to manage the scheme.
• Liability. Unredeemed points are a promise you have made.

Whether points, tiers or both increase repeat purchases is a question for your data, not for a vendor's average. Count the reward as a variable cost of the order. Our convention is to judge it on contribution margin before marketing: revenue minus the variable costs that grow with every order, which are cost of goods, shipping, payment fees, returns and discounts. See contribution margin for the definition. Points and tier benefits belong in that cost line. Sources differ on which costs they include, so write your list down.

A discount reduces price, and that is why it hurts. An ecommerce analytics vendor argues that a discount shrinks profit margins faster than it reduces revenue⁴. The mechanism is simple arithmetic: the whole price cut comes off the margin, while the cost of goods, shipping and fees on the order stay the same.

There is a behavioural cost too. A code in every email teaches people to wait for the next one. A customer who would have paid full price learns that full price is optional. For the discount rate, meaning the amount of the cut as a share of the original price, the question is not "what is ours?" but "which orders needed it?".

Discounts have a place: clearing stock, a first order, a lapsed customer you are trying to win back. Treat them as a tool with a test, not a default.

You have two honest options.

1. Holdout. Randomly keep a slice of eligible customers out of the programme, or out of the discount, for a set window. Compare repeat purchase rate and margin per customer with the group that got it. This is a holdout test. Random assignment removes the selection effect.
2. Matched groups. If you cannot hold anyone out, compare members with non-members who look alike before joining: similar order history, category, first-order date and value. This is weaker, because you can only match on what you can see. Say so when you report it.

Whichever you use, decide the window and the metric in advance. Count margin, not revenue. A programme that raises repeat rate and lowers contribution margin per customer is not a win. Add the cost of the rewards.

Our view: if you do not want to hold anyone out, the matched comparison is the minimum. A before-and-after chart of members alone is not enough.

Sometimes the right answer is "neither yet". Look for these signs:

• Your repeat purchase rate is already healthy for your category, and the main issue is first orders.
• Your product is bought rarely, so points would take years to matter.
• Your margins are thin, and rewards would eat what is left.
• You cannot afford to run a proper comparison, and you would be guessing.

In those cases, fix the basics first: delivery, product quality, post-purchase email. Cheaper levers often move repeat rate before a scheme does.