Loyalty Programs for Consumer Goods That Actually Move Profit

Loyalty Programs for Consumer Goods That Actually Move Profit

Most FMCG operators run loyalty programs that reward the customer who would have bought anyway. McKinsey's research on loyalty economics is direct about it: spend on loyalty programs rarely produces incremental behaviour and instead subsidises actions customers were already inclined to take. That is a precise, expensive way to describe a profit leak running through the consumer-goods category. A points-per-dollar program imported from aviation or hotel economics, deployed unchanged on a six-dollar product with a 30-day repurchase cycle, taxes margin while pretending to grow share.

This is not a marketing problem. It is a unit-economics problem dressed up as a CRM strategy.

The Subsidy Hidden Inside Every Points-Per-Dollar Program

The case for the standard FMCG loyalty program rests on three assumptions, all of them false at six-dollar unit prices. The first assumption is that points-per-dollar redemptions reward incremental spend. The second is that enrolment growth proves the program is working. The third is that members who redeem rewards are more loyal than non-members. McKinsey's McKinsey loyalty value work tears each one apart with a single observation: loyalty spending typically subsidises behaviour customers were already going to take, then claims credit for it on a dashboard.

The CMSWire team frames it more bluntly. The points-equal-preference assumption is what they call the Loyalty illusion. A buyer joining your program because they were already buying weekly is not new behaviour. Their points balance grows. Your reporting glows. Your contribution margin shrinks by the cost of every redemption that would have happened without the program. Across the FMCG operators I have worked with, the cannibalisation pattern is consistent: more than 70 percent of redemptions come from buyers whose pre-program purchase frequency was already at or above the brand average. They were locked in. The program just gave them a cheaper price.

Upside's analysis of the same pattern is sharper still. Every discount handed to a customer who would have bought at full price is profit left on the table, because Upside incremental sales are the only kind that justify the program's cost. Not enrolment. Not redemption rate. Not member-vs-non-member spend ratios that conflate selection bias with causation. The whole metric stack that vendors sell to FMCG brand teams is engineered to hide the cannibalisation problem because surfacing it would kill the contract.

The math is brutal at the unit-economic level FMCG operates at. A typical CPG SKU runs a contribution margin of 25 to 35 percent on a six-dollar retail price, which is roughly $1.50 to $2.10 per unit before any marketing cost. A single free-product redemption erases the margin on roughly three months of that buyer's purchases. If 70 percent of redemptions go to already-loyal buyers, you are paying full margin cost to retain customers who needed no retention. This is the structural failure the standard program never confronts. It runs on aviation and hotel program logic, where ticket prices are high enough that a free flight rewards meaningful spend. On a product where the unit price is the cost of a coffee, the same logic destroys the P&L.

The Repeat Purchase Engine

The Repeat Purchase Engine is the system I deploy with FMCG operators in the $1M to $10M revenue band who want a loyalty program that does not subsidise its own non-incremental cost. It rejects the points-per-dollar default and replaces it with a single accountable metric: incremental purchase frequency among buyers who were not already buying weekly. The program rewards behaviour change, not behaviour confirmation. A buyer who bought once a quarter and now buys monthly is the program's only valid output. Everyone else is noise the dashboard pretends is a result.

The engine has three components. First, a buyer segmentation layer that classifies every member by their pre-enrolment purchase frequency, drawn from DTC orders, on-pack QR registrations, or matched retailer loyalty data. Second, a tiered reward structure where the largest rewards are reserved for casual buyers who increase their purchase frequency, not for heavy buyers who maintain it. Third, an always-on holdout cohort, a randomly selected non-enrolled control group that lets you measure whether the program produced incremental behaviour or simply rewarded existing behaviour at a discount.

I have seen what happens when an FMCG brand swaps the standard program for this engine. One Australian pantry-staples brand running a 1-point-per-dollar program with 80,000 members was redeeming roughly 14 percent of issued points, in line with the LoyaltyLion redemption industry average. After segmentation, they discovered that 73 percent of redemptions came from members already buying at or above category average. The program was not a loyalty engine. It was a discount channel for their best customers. They restructured the rewards around frequency uplift, halved the points yield for top-decile buyers, and tripled it for casual buyers who shortened their repurchase cycle by 30 percent or more. Twelve months later, contribution margin per member rose by 18 percent and incremental purchase frequency among the casual segment moved from quarterly to every six weeks. They did not add features. They removed the subsidy.

McKinsey's work on the McKinsey eight levers of modern loyalty design reaches the same destination from a different angle: programs win when they go beyond points and redemption mechanics, and they fail when they do not. The Repeat Purchase Engine treats the points-per-dollar mechanic as a legacy artifact, not a starting point. The same insight runs through LoyaltyPoint FMCG design guidance: low-ticket, high-frequency products demand mechanics that look nothing like the airline-mile templates the industry copies by default.

Phase 1: Audit the Existing Program (Days 1-30)

Phase 1 is a forensic exercise, not a strategy exercise. The job is to put a number on the cannibalisation that has been hiding inside your loyalty P&L. Most FMCG operators avoid this because the answer is humiliating. Run it anyway. The number is the precondition for every later decision.

Pull twelve months of redemption data. For every redemption, attach the member's pre-enrolment purchase frequency. If you do not have pre-enrolment data because the member joined before the analysis window, use their first 90 days of post-enrolment frequency as a proxy, with a note that this understates cannibalisation because the program may have already biased their behaviour. Now segment members into four bands by frequency: heavy (weekly or more), medium (fortnightly to monthly), light (monthly to quarterly), and lapsed (less than quarterly). Tag every redemption to the band the buyer occupied at enrolment.

Calculate the cannibalisation ratio: redemptions from heavy and medium buyers divided by total redemptions. The number you produce is the percentage of program cost that subsidised existing behaviour. In every audit I have run on FMCG accounts in the $1M to $10M band, the ratio sits between 65 and 82 percent. The headline diagnostic in Five failure signs is that members redeem rewards but never change purchase frequency, and the audit makes that signal quantifiable. If your ratio is above 50 percent, your program is a discount channel for your best customers. If it is above 70 percent, the program is a margin hole.

Now produce a per-member contribution-margin report. For each member, calculate their twelve-month spend, the contribution margin on that spend, and the cost of every redemption they triggered. Subtract redemption cost from contribution margin. Sort descending. The members at the top are profitable. The members at the bottom, the ones whose redemption cost exceeds their incremental contribution, are the program's structural loss-makers. In the audits I run, between 18 and 30 percent of members fall into the loss-maker tier. They are not bad customers. They are good customers the program is overpaying.

Document the program's enrolment-source mix. If 60 percent or more of enrolments came through prize-draw mechanics or competition entries, expect cannibalisation to skew higher and verified-purchase rates to skew lower. Tag those members for re-verification before the rebuild. The audit finishes with four numbers on a single page: cannibalisation ratio, member contribution-margin distribution, loss-maker percentage, and enrolment-source mix. Send that page to your CFO. The conversation that follows is the moment the program either gets rebuilt or gets killed, and either is a better outcome than the slow margin leak the brand has been running.

Phase 2: Rebuild Around Incrementality (Month 2-3)

Once the audit is done, the rebuild is mechanical. The Repeat Purchase Engine reorganises rewards around a single qualifying behaviour: a buyer must demonstrate frequency uplift, category expansion, or basket expansion to access the largest rewards. Heavy buyers who maintain their existing pattern earn baseline rewards only. The economic logic is simple. Reward the change you want to cause. Stop paying full price to confirm behaviour you already had.

Restructure your reward tiers around three qualifying actions. Tier 1 rewards are for casual or lapsed buyers who shorten their repurchase cycle by at least 25 percent over a 90-day window. Tier 2 rewards are for buyers who add a SKU from a category they had not previously purchased. Tier 3 rewards are for buyers who increase basket size by at least 20 percent on a single order. Heavy buyers who do none of these three earn baseline points only, and the baseline is set well below industry-default 1-point-per-dollar levels. Roughly 70 percent of program budget should now flow to behaviour-change rewards, with the remaining 30 percent on baseline retention.

Communicate the change as a benefit, not a cut. Existing heavy buyers should see the messaging frame the rebuild as a new tier they qualify for through behaviour, not as a downgrade. Run a 60-day amnesty period where existing point balances retain their old conversion rate, after which they convert to the new structure. Antavo's Antavo loyalty census work shows that nine in ten loyalty operators report positive ROI on their programs, but the methodology rarely separates incremental from cannibalised spend, which is exactly the gap the rebuild closes. You are not running a worse program. You are running the same program with a cleaner accounting of what counts as a result.

Build the rewards to fit FMCG unit economics, not airline economics. The rewards that work on a six-dollar product are sample-sized: a free travel pack of a complementary SKU, a recipe booklet built around your products, exclusive access to a limited-edition variant, or shipping credit for the DTC store. Avoid free full-size products as default rewards because they erase three months of contribution margin per redemption. The LoyaltyXpert CPG guide on CPG-specific loyalty design lays out the unit-economic constraint clearly: rewards must cost less than the marginal contribution from one additional purchase, or the program is structurally unprofitable.

Set up the data plumbing for the rebuild before you launch it. Every reward issued must carry a tag indicating which qualifying behaviour earned it, so you can later measure whether the program is producing the behaviour change it claims. Without that tagging, you are running a black box again, and the new program will accumulate the same dashboard sins as the old one within six months.

Phase 3: Install the Holdout Cohort (Quarter 2 Onward)

Phase 3 is the only phase that produces credible evidence the program works. Without it, every metric the program reports is comparing the program against itself. The fix is structural: a randomly selected cohort of buyers who never get enrolled, never see the program, and never receive its rewards. Their behaviour over time is the counterfactual. Subtract their average purchase frequency from your members' average, and the difference is the only number that proves incremental value.

Randomly assign 5 to 10 percent of newly acquired buyers to a permanent holdout. Suppress all program communications to them. Do not show them program upsells in the DTC store. Do not include them in retailer-loyalty enrolment partnerships. Track their purchase frequency, basket size, and category mix in the same data structure as enrolled members. The cleanest deployment runs through the DTC store and matched retailer-loyalty cohorts where you can deterministically suppress treatment. On-pack QR registration is harder to randomise, so apply the holdout to the channels you can fully control first.

Report two metrics monthly: enrolled-member purchase frequency minus holdout-cohort purchase frequency, and enrolled-member contribution margin minus holdout-cohort contribution margin. Both numbers are net of program cost. The first proves the program changed behaviour. The second proves the behaviour change was worth the rewards you paid for it. If either number is flat or negative for two consecutive quarters, the program is not working at the structural level, and Phase 2 needs another rebuild pass.

The holdout cohort is the single piece of architecture the standard FMCG loyalty program never includes, because it would falsify the dashboard the vendor sold the brand on. Antavo's Antavo statistics industry-cut benchmarks show that member-vs-non-member spend ratios are reported as a primary success metric across CPG and retail. That comparison is selection bias dressed up as causality. Members spend more because they were always going to spend more, which is why they joined. The holdout corrects for this by removing self-selection from the comparison.

Maintain the holdout permanently. Do not rotate buyers in and out of it. Do not enrol successful holdouts in the program once they hit a frequency threshold. The cohort needs to remain comparable across years to detect changes in program performance over time. Treat it the way pharmaceutical companies treat a control group: contaminate it and you have lost the only evidence base that lets you know whether the program is doing anything at all.

The New North Star Metric: Incremental Purchase Frequency

The Repeat Purchase Engine produces one number that should sit on the brand manager's monthly P&L, and it replaces the four or five vanity metrics the standard program reports. That number is incremental purchase frequency: the difference in average annual purchase frequency between enrolled members and the holdout cohort, measured by buyer segment.

Stop reporting enrolment growth as a success metric. A million enrolments produces zero profit if all of them are heavy buyers redeeming rewards on purchases they were already making. Stop reporting redemption rate as a success metric. A 25 percent redemption rate is bad if 73 percent of those redemptions came from buyers who needed no incentive. Stop reporting member-vs-non-member spend ratio. The comparison is structurally wrong because it does not control for self-selection.

Report incremental purchase frequency by segment, broken into casual, lapsed, and category-expansion cohorts. Report contribution margin per member, net of program cost, against the same cohort breakdown. Report the cannibalisation ratio quarterly, as the leading indicator of program health. When the cannibalisation ratio drops below 35 percent, the program is operating as a loyalty engine. When it stays above 60 percent, the program is operating as a discount channel.

The shift is from a program measured on its own activity to a program measured on its causal effect on the buyer. That is the only frame in which an FMCG loyalty investment makes financial sense at six-dollar unit prices and 25 to 35 percent contribution margins. Every other frame is a slow leak the dashboard hides until the CFO asks why the loyalty line item grew 40 percent and category share moved sideways. Build the engine. Run the holdout. Report the right number. The program either earns its place on the P&L or comes off it, and either decision is a better one than the program you have been running.