A loyalty scorecard can climb all year while the business beneath it stands still.
That is not a reporting accident. It is what happens when a dashboard measures reach and calls it growth. Enrollment goes up, and so does member share of sales, the portion of revenue rung up with a member ID attached. The program takes the credit at the quarterly review, and the question finance cares about most goes unasked: how much of that revenue would have come in anyway?
For years, a regular has ordered the same oat latte on her way to work. One February morning the barista asks if she has the app, and she scans in before the milk is steamed. Her mornings do not change. The café’s report does, even though she was buying every one of those lattes before she joined. Multiply her by a few thousand regulars, and the report that reaches finance describes a program in perfect health.
Reach and visits moved in opposite directions at Starbucks
The same split shows up at scale. In its October 2024 results, Starbucks reported that U.S. Rewards members active in the past 90 days rose 4% in a year. That is an activity window: any member with activity inside those three months stays in the count, however rarely she visits. In the same quarter, U.S. comparable transactions fell 10%. The audience grew while the visits fell.
A few months later the member count dropped out of the earnings release, and none of the six releases since has carried it. The releases give no reason, and the pattern is worth copying anyway: lead the program report with the visits and margin finance already tracks, and put the member count beside them as context.
A member sale shows what the program can see, not what it made
Member share of sales tells a team how much of the business carries a member ID. How much of the business exists because of the program is a separate question, and the share cannot answer it. When the regular scanned in, her spend moved from one column to another, and the café did not sell one more latte.
Loyal customers also tend to join first, and many were already shopping more before they enrolled, so part of the gap between members and non-members was there before anyone signed up. Researchers call this a selection effect. Every member premium blends two effects: what membership does to a customer, and which customers choose to join. Both are worth having. Only one of them is the program’s doing.
A 2007 study of Dutch grocery shoppers measured how far apart the two can sit. Membership did raise share of wallet, the portion of a household’s grocery spending that goes to one chain, but by seven times less than a naive model showed. A naive model is a straight comparison of members with non-members that ignores who chose to join. For the analyst who reports to finance, that makes the member premium an upper bound on what the program can claim. Label it that way until a test measures the part membership caused.
Members say the program works, and a holdout checks whether it does
Surveys cannot pull the two effects apart, because they record what members believe about themselves. Bond, a customer engagement firm, surveyed more than 20,000 U.S. program members in 2026: 85% said a program makes them more likely to keep doing business with a brand, and 73% said they spend more as a result. Loyalty Juggernaut frames the real test as a question to ask of every offer: which offers are producing incremental behavior, and which are subsidizing existing behavior? A survey records what members say, so it cannot answer that. A holdout records what they do.
A holdout is a random group of eligible customers kept out of an offer, or out of the program’s marketing, while everyone else receives it as usual. Because chance alone decides who is held out, the two groups start alike, and any gap in spending at the end is what the offer caused. Analysts call that gap incremental: revenue that would not exist if the offer had never run.
Timing matters as much as the split. A study of 322 public companies found that a new program lifted sales and gross profit for at least three years, and that profit registered only from the second quarter after launch. A test that ends after one quarter reads the fast half of the result. Agree the test length with finance before the offer goes out, so nobody ends it on the first good week.
A holdout also has limits, and a measurement team earns trust by naming them. A program cannot hold members out of benefits their tier already promises, so the method fits offers and marketing, not the core terms of membership. A small segment rarely holds enough members in each group to separate a real gap from noise. Offers leak, too, when a held-out member sees a friend’s promotion or a partner’s ad. Where a clean holdout is impossible, a staggered rollout that launches in one market before another, or a comparison of matched markets, gives the next-best read. For a change that plays out over years, such as a new tier structure, the answer is a small program-level control group kept for the long run, checked against every shorter test the program runs.
Finance reads the loyalty scorecard as a cost report
Every reward is margin handed back to a customer. Every point outstanding sits on the balance sheet as an obligation until the member redeems it or it expires. That obligation is the points liability, a form of deferred revenue. Breakage is the share of points the program expects never to see redeemed. Both numbers rest on the long-run redemption rate the program assumes. Change that one assumption and the balance sheet moves, even when no member changes what she buys.
The growth those rewards are meant to buy is real only when a test shows it. Sooner or later, finance asks which of these sales would have happened anyway. A program that cannot answer will keep defending its budget with member sales, and those will keep rising whether the program works or not.
Bring the answer to that meeting before the question comes: the holdout result, the full cost behind it, and the liability that grew alongside it. That is the evidence that loyalty works as an enterprise growth engine, and a member count cannot supply it.
Four layers sort the KPIs by the question each can answer
Sort every KPI by the question it can answer, and the scorecard stops arguing with itself. Reach shows who the program can see, starting with enrollment and member share of sales. Engagement shows whether members keep coming back, through measures such as the active rate and churn. Economics shows what the program owes and what members are worth, including customer lifetime value on margin. Causality shows what the program changed, and no other layer can.
Enrolling a regular lifts three rows of the full table below the day she scans in. Only the causality layer shows whether any of it changed what she buys.
The layers hold in every industry, while the KPI at the top of the page changes with what the business sells. A hotel group leads with member share of room nights and direct booking. A restaurant chain leads with the share of transactions carrying a member ID, next to same-shop transactions, the visits at shops open long enough to compare. Dutch Bros shows why both belong on one page: in the second quarter of 2026, its Rewards share of transactions rose while same-shop transaction growth slowed. Pick the lead KPI from what the business sells, and set a visit or margin measure beside it.
A worked holdout puts a price on growth
The worked example below runs one illustrative test from start to finish. A café chain keeps a random tenth of its eligible customers out of the program’s offers for two quarters, then compares spend per customer in the two groups. The gap becomes incremental margin once gross margin is applied. Set the full cost, rewards included, against that margin, and finance gets the two numbers it asks for: 80 cents of cost for each dollar of margin the program caused, and a 25% return.
The same data can be divided in ways that flatter the program. Compare the two groups’ totals, and the lift reads 872%, because the offer group is nine times larger. Divide the cost by total sales, and growth looks almost free.
Show finance the per-customer math, and write down every divisor before the test starts.
Six rules turn a KPI list into a scorecard
- Label reach as reach. Enrollment and member share of sales describe coverage. Give them their own line next to total transactions, so nobody reads a rising share as growth.
- Fix the definition of “active” before counting. Choose the action that counts, such as a purchase, and the window, such as 90 days. Then hold both still. Change either one and the trend line moves while the customers stay exactly where they were.
- Put the points owed next to the sales. Show points outstanding, liability per member, the redemption rate the liability assumes, and reward cost per member beside member sales. A program that reports what it sold without what it owes is telling half the story. The points liability accounting guide covers the accounting.
- Hold out a random group from the start. Keep a random slice of eligible customers out of each offer, and out of the program’s marketing where the rules allow, then compare spend and margin. Run the test long enough for profit to show. A holdout chosen by hand is not a test. It is an argument.
- Price the growth. Divide the program’s full cost, rewards included, by the incremental margin the holdout shows before reward costs, so no reward is counted twice. That ratio is the number to defend in a budget meeting. The loyalty program ROI guide walks through each input.
- Ask every new member why she joined. One question at enrollment, whether someone referred her, lets the program count referral revenue from its own records instead of a survey score. Reichheld’s earned growth rate is built on that answer: revenue growth from returning customers and their referrals.
The fourth rule carries the other five. Without a holdout, the scorecard describes the program. With one, it explains it.
GRAVTY® tests the offer before the budget moves
Within weeks of her joining, GRAVTY’s patented Multi-Dimensional Behavior Tracking had drawn a settled habit on the regular’s record. She came in at the same hour every weekday and redeemed her first reward before any offer reached her.
Agentic AI Compass reads the whole program. Its AI agents work like a team of expert analysts, reading KPIs alongside sentiment and benchmarks and flagging anomalies as they appear. They simulate an offer’s outcome before launch, and every recommendation stays explainable and auditable.
Say the quarter’s member sales climb while visits per member hold flat. Compass traces the climb to regulars who enrolled at the counter. It recommends the next offer for members whose weekends are still open, and it simulates that offer before the budget moves. The weekday regular keeps paying full price, as she always has. The loyalty team then runs the weekend offer against a holdout, so finance gets a recommendation it can audit and a lift it can count.
The weekday latte was always hers
On a wet Saturday in April, the regular walks four blocks for a double-points offer. It is the first weekend visit on her record, and the report logs it as one more member transaction, identical to the weekday ones.
It is not identical. Her weekdays measured the program’s reach. That Saturday is what a holdout exists to find.
The column the report puts her in is invisible to her. What she notices is whether the café knows her well enough to save its offers for the parts of her week that are still open.
A scorecard built on reach will look healthiest in the program customers already love. The harder number, and the one worth taking to finance, is the one that shows what the program changed.
The weekday latte was always hers. The Saturday one belongs to the program.
The numbers behind loyalty program KPIs
These are the tables behind the essay. Each figure links to the filing, study, or release it comes from, and the café test uses illustrative figures.
Seventeen KPIs in four layers
Every KPI below is worth tracking. Each one also has a way to flatter the program.
| Layer | KPI | Formula | What it proves | The trap |
|---|---|---|---|---|
| Reach | Enrollment rate | New members ÷ eligible customers | The invitation to join works | Sign-ups bought with a one-time discount |
| Reach | Member share of sales | Member sales ÷ total sales | How much of the business the program can see | It rises every time a regular scans in, even when her order stays the same |
| Engagement | Active rate | Members with a purchase in the window ÷ all members | The program is part of a routine | A new window or a new action changes the answer |
| Engagement | Purchase frequency and average order value | Member transactions ÷ active members; member sales ÷ member transactions | Habit and spend | Loyal buyers tend to join first |
| Engagement | Repeat purchase rate | Members with two or more purchases in the window ÷ members with at least one | Members come back within the window | Regulars repeat with or without the program |
| Engagement | Retention rate | Members active in both periods ÷ members active in the first | Members keep coming back | Many loyal members would have stayed anyway |
| Engagement | Churn rate | 1 minus the retention rate for the same two periods | How fast active members stop buying | The enrolled count hides it, since lapsed members stay on the list |
| Engagement | Redemption rate | Points redeemed ÷ points issued | Members find the rewards worth reaching | Its long-run version, the expected redemption rate, sets the liability estimate |
| Engagement | Net Promoter Score | Share of promoters minus share of detractors | Stated willingness to recommend | Self-reported and unaudited |
| Economics | Points liability | Points outstanding × expected redemption rate × standalone selling price per point | What the program owes | The redemption assumption moves it |
| Economics | Breakage or expiration rate | Breakage: 1 minus the expected redemption rate. Expiration: points expired ÷ points issued | The share of points the program expects to keep | Rising breakage shrinks the liability and can mark members who gave up |
| Economics | Tier mix and reward cost per tier | Members in a tier ÷ all members; reward and benefit cost in a tier ÷ members in that tier | Where the reward budget goes | A tier’s cost is easy to see; its lift needs a holdout inside the tier |
| Economics | Customer lifetime value on margin | Discounted member margin after rewards over the expected relationship | Whether members are worth what they cost | Comparing it with non-members measures selection |
| Economics | Earned growth | Revenue growth from returning customers and their referrals | Growth that customers brought in | It needs every new customer’s reason for joining, and a holdout still decides cause |
| Causality | Incremental lift | (Offer group spend per customer ÷ holdout spend per customer) minus 1 | What the offer caused | A holdout chosen by hand or ended early |
| Causality | Cost per incremental margin dollar | Full program cost, rewards included ÷ incremental margin before reward costs | What growth costs | Dividing by total member sales |
| Causality | Loyalty program ROI | (Incremental margin before reward costs minus full program cost, rewards included) ÷ full program cost | Whether the program pays back | Counting all member margin as incremental |
A worked holdout, step by step
A control group test turns the causality rows into one number. The figures in this example are illustrative. A café chain has 100,000 eligible customers. It keeps a random 10% out of the program’s offers and marketing for two quarters and runs the program as usual for the other 90%.
| Step | Formula | Illustrative result |
|---|---|---|
| Random split | 90% get offers, 10% are held out | 90,000 and 10,000 customers |
| Spend per customer | Group spend ÷ group size | Offer group: $9,720,000 ÷ 90,000 = $108.00. Holdout: $1,000,000 ÷ 10,000 = $100.00 |
| Incremental lift | ($108.00 ÷ $100.00) minus 1 | 8% |
| Incremental spend | ($108.00 minus $100.00) × 90,000 | $720,000 |
| Incremental margin before reward costs | $720,000 × 40% gross margin | $288,000 |
| Full program cost, rewards included | $172,800 in rewards plus $57,600 for the platform and staff | $230,400 |
| Cost per incremental margin dollar | $230,400 ÷ $288,000 | $0.80 |
| Loyalty program ROI | ($288,000 minus $230,400) ÷ $230,400 | 25% |
Each dollar of margin the program caused cost 80 cents, a 25% return. Divide the group totals and the same test shows a lift of 872%, because the offer group is nine times larger. Divide the cost by the offer group’s $9,720,000 in sales and growth looks almost free, at 2.4 cents a dollar.
Starbucks dropped the member count and kept the visits
| Starbucks release | What it reported | How to read it |
|---|---|---|
| October 2024 | 33.8 million U.S. Rewards members active in the past 90 days, up 4%; U.S. comparable transactions down 10% | Reach and visits moved apart |
| January 2025 | 34.6 million members, up 1%; U.S. comparable transactions down 8% | The last release to carry the member count |
| April 2025 through July 2026 | Six releases report transactions and leave the member count out | The headline moved to the business |
| Quarter to December 2025 | U.S. comparable transactions up 3% | Visits turned up before the relaunch |
| Quarter to March 2026 | A reimagined Starbucks Rewards launched with three levels of membership | A program change after the turn |
| Quarter to June 2026 | U.S. comparable transactions up 4.2% | Growth counted in visits |
Research separates the member premium from the program’s effect
| Source | What it found | What it means for the scorecard |
|---|---|---|
| Leenheer, van Heerde, Bijmolt, and Smidts, March 2007 | In a panel of Dutch households covering all seven grocery loyalty programs, membership raised share of wallet by an amount seven times smaller than a naive model showed | Members start ahead of non-members |
| Bond, a customer engagement firm, June 2026 | Of 20,591 U.S. loyalty program members surveyed, 85% said a program makes them more likely to keep doing business with a brand, and 73% said they spend more as a result | Stated intent |
| A study of 322 public companies, April 2019 | Programs launched between 2000 and 2015 lifted sales and gross profit for at least three years; profit registered from the second quarter after launch, well behind sales | A short test reads the fast half of the result |
| Fred Reichheld, creator of the Net Promoter Score, and two Bain colleagues, Harvard Business Review, November 2021 | Introduced the earned growth rate: revenue growth from returning customers and their referrals, drawn from accounting results | It needs each new customer’s reason for joining |
The KPI each industry leads with
| Industry | KPIs to lead with | What a public filing shows | How to read it |
|---|---|---|---|
| Hotel | Member share of room nights, direct booking rate, tier mix and cost per tier | IHG, August 2026: members booked 67% of room nights, spend “~20% more,” and are “around 10x more likely to book direct” | Each gap blends what membership does with who joins |
| Restaurant | Rewards share of transactions, active members in a stated window, same-shop transactions | Dutch Bros, August 2026: Dutch Rewards carried 73% of transactions, up from 72%, while same-shop transactions grew 1.7%, down from 3.7%, both against the second quarter of 2025 | The company reads the share “as an indicator of customer loyalty adoption” of its app; same-shop transactions track growth |
| Retail | Member share of sales, repeat purchase rate, average order value, breakage | Starbucks, October 2024: 90-day active U.S. members up 4%, U.S. comparable transactions down 10% | A member count can grow while visits fall; the retail loyalty programs guide compares Starbucks Rewards with other designs |
Frequently asked questions
What are the five most important loyalty program KPIs?
The five most important loyalty program KPIs are retention rate, repeat purchase rate, redemption rate, customer lifetime value, and incremental ROI. Retention and repeat purchase show whether members keep coming back. Redemption rate shows whether the rewards are worth reaching, and it drives the points liability finance carries. Customer lifetime value on margin shows whether members are worth what they cost. ROI, measured against a random holdout, is the one of the five that shows what the program caused, so it belongs at the top of any report to finance.
How do you measure the success of a loyalty program?
Measure a loyalty program’s success with a holdout: compare members who got an offer with a random group who got none. Count the difference in spend and in margin before reward costs, then set the program’s full cost, rewards included, against that margin. Agree the test length with finance before launch, and run it long enough for profit to show. In a study of 322 companies, profit from a new program lagged well behind sales. To measure the whole program, use a program-level holdout or a staggered launch.
How do you calculate loyalty program ROI?
Loyalty program ROI is the margin the program caused, minus its full cost, divided by that cost. The margin comes from a random holdout and is counted before reward costs. The cost includes rewards. For example, $288,000 of incremental margin against $230,400 of cost gives an ROI of 25%. Counting all member margin credits the program with sales regulars would have made anyway. The loyalty program ROI guide covers holdout design.
Why do loyalty members spend more than non-members?
Loyalty members spend more partly because of the program and partly because loyal customers tend to join first. A 2007 study of Dutch grocery shoppers found that membership did raise share of wallet, by an amount seven times smaller than a naive model suggested. Report the member premium to finance as an upper bound on the program’s effect until a holdout shows how much of it membership caused.
What is a good redemption rate for a loyalty program?
A good redemption rate matches the redemption assumption behind the program’s points liability and holds steady or rises among active members. A rising rate means members find the rewards worth reaching, and it raises the liability per point outstanding. A falling rate can mean rewards feel out of reach, or that new members are still earning toward their first. One public figure dates from 2011: COLLOQUY and SWIFT EXCHANGE estimated that at least a third of the perceived value of US points and miles goes unredeemed.
