Members buy more in every dataset ever published, because your best customers join first. Loyalz runs the programme and measures whether it actually caused anything.
A loyalty program is a structured incentive to buy again — points, tiers, benefits, rewards. Whether yours works is a harder question than it looks, because the usual evidence does not survive scrutiny: your best customers join first, so members outperform non-members whether or not the programme changed anyone's behaviour. The only way to know is to withhold it from a randomly chosen group and compare.
This is the chart the whole category sells on, and it is real. It is also uninformative, because nobody assigned those customers to those groups — they chose.
Every member-versus-non-member ratio you have ever been shown is consistent with a programme that does nothing at all. We sell a loyalty product and we are telling you this, because the alternative is selling you a number we know is uninformative.
An order-level split leaks — the same person lands in both arms and the effect washes out.
A customer who moves arms mid-experiment belongs to neither. Store the arm, the timestamp and the seed.
Withholding points while leaving tier benefits running measures a mechanic, not a programme.
One cycle shows whether you pulled a purchase forward. Two shows whether you created one.
Metric, window and cut-offs committed before the first assignment. Deciding after you look is how a null becomes a positive.
Dividing by actives quietly drops everyone the programme failed to reach — the group it was supposed to affect.
Revenue per customer is skewed, so this needs more customers than instinct suggests. Per arm, two-sided, 80% power:
Scroll the table sideways to see every column →
The instinct is to hold out 5 or 10% because withholding the programme costs revenue. That makes the test more expensive, not less: a 10/90 split needs about 2.8× the total customers of an even split, because the smaller arm limits the comparison.
Worked example of the shape the category never draws. A reward redeemed by someone who would have repurchased anyway is not marketing spend — it is margin handed over for nothing, and the resulting order still gets counted as loyalty-driven revenue.
01 STARTER
02 EXPLORER
03 ELEVATED
04 PREMIER
05 ICON
Illustrative layout. Points, cashback, tiers with experience levels, quests and benefit collections — with reward cost sitting on the same screen as the revenue it is supposed to have produced, which is the only arrangement that lets you catch a programme paying for its own results.
RFM segmentation tells you who is worth keeping before you spend anything on keeping them. Recency, frequency and monetary value are the three facts about a customer that predict the next order, and they need no model to compute.
Bring your order history. We compute the effect size you could actually detect, before you commit to withholding anything from anyone.
Book the twenty minutes →