Loyalty Programmes
Date: 2026-08-16
Rewarding repeat behaviour. The mechanics are easy and the economics rarely get checked — most of the reward goes to customers who were already loyal, which is a margin transfer rather than a growth lever.
What it is
A loyalty programme gives customers value in exchange for repeat purchase — points, tiers, cashback, or member pricing.
The structural problem
Rewards go disproportionately to your best customers, who were buying anyway.
customer base reward earned behaviour change
frequent buyers 20% 65% little
occasional 50% 30% some
one-time 30% 5% little
Most of the cost lands where it changes nothing. That’s not a badly designed programme — it’s inherent, because rewards scale with spend and spend scales with existing loyalty.
Which makes the honest question not “does the programme drive repeat purchase” but “does it drive enough incremental repeat purchase to cover what we give away to people who’d have bought anyway” — Incrementality Testing.
The margin arithmetic
A points scheme returning 5% of spend:
£750,000 annual revenue
× 5% reward rate = £37,500 of value issued
× ~70% redemption = £26,250 actual cost
annual contribution £225,000
cost as share of contribution 12%
A 5% reward rate costs 12% of contribution on a 30% margin. To break even it must generate roughly 12% more contribution — around 1,750 additional orders a year — Break-Even Analysis, Contribution Margin.
That’s a demanding target, and the calculation is rarely done before launch.
What makes them work
Where they do earn their keep, the mechanism is usually psychological rather than economic:
- Tiers create a goal. Progress towards a threshold motivates more than the reward’s cash value, and tiers cost less than equivalent discounting
- Points have a break in liability. Unredeemed points cost nothing, which is why redemption rate is the number that decides the economics
- Membership changes the default. A member checks your site first, which is worth more than the reward
- Data. A loyalty identity solves guest-checkout identity problems, which improves Repeat Purchase Rate measurement and every lifecycle flow downstream — sometimes the strongest argument for one
That last point is underrated: a programme that gets customers to identify themselves fixes measurement problems that no analytics work can — Identity Stitching.
What makes them fail
- Rewards indistinguishable from a permanent discount. 5% back on everything is a 5% price cut with extra steps, and it trains nobody
- Thresholds too distant to motivate
- Complexity. If a customer can’t state what they’ll get, it isn’t influencing them
- No exit. Loyalty liabilities accumulate on the balance sheet and are politically hard to withdraw once customers expect them
- Rewarding volume rather than margin, so the programme subsidises low-margin buying — Basket Composition
Testing one
Hard, because you can’t easily randomise membership — but not impossible:
- Hold out a random slice from launch communications and compare enrolment and behaviour
- Geo test a tier or a reward change — Geo Holdout Tests
- Test changes to an existing programme rather than its existence, which is far more tractable
- Compare enrolled and non-enrolled with heavy caveats. Enrolment is self-selected — enrolled customers were already better — so the comparison is Selection Bias rather than an effect
In plain terms: loyalty programmes are easy to launch, hard to measure, and impossible to withdraw. Work out the break-even before committing, and prefer mechanics that cost you only when the customer does something they weren’t going to do anyway.