Customer Loyalty: the Mistakes That Cost EBITDA and the Method That Lifts Guest LTV

1. Where does the points program actually get paid?

A redeemed point is paid out of the dish's contribution margin, not the marketing budget, and that single accounting line decides whether your program builds cash or drains it.

In the three-location group opening this brief, above 5 million dollars a year with 11,400 enrollees in fourteen months, points were budgeted upstairs and discounted downstairs, so the marketing report celebrated while the P&L bled. Treat every point as a VARIABLE COST OF SALE and subject it to the same scrutiny you apply to food cost, with the hard ceiling of 32% per dish, because an 8% redemption on a dish already running at 30% leaves barely 62 cents of each dollar to cover payroll, rent and utilities. The benchmark exists: according to Paytronix (Loyalty Trends Report 2024), operators in the 90th percentile draw more than 37% of their transactions from loyalty members.

2. The metric is not sign-ups: it is frequency by cohort

Counting enrollees measures nothing, because a large database with redemption concentrated among guests who were already returning is a subsidy dressed up as strategy. Measure frequency by first-visit cohort and twelve-month guest value, and ask the uncomfortable question before signing any platform renewal: of those 11,400, how many would have come back anyway without the discount? The cohort that matters is the guest who came twice and vanished, not your Thursday regular. Paytronix data (2024) sets the yardstick: loyalty members buy more often than non-members, so if your active base is nowhere near that mark, you own a mailing list, not a program. Transactional penetration rules; the size of the list never does.

3. Data, content, incentive: that order and no other

The correct order of levers is data first, content second, incentive last, and the classic mistake inverts it: coupon on Monday, content when time allows, data never. The entry door today is content, not discount, because according to Restroworks (Restaurant Social Media Statistics 2025), 72% of people use social media to research restaurants, and per the TouchBistro 2025 Diner Trends Report, 57% of millennials decide where to eat based on what they see on social. Diego F. Parra says the same thing in the kitchen and in the boardroom: a discount buys one visit, the brand buys a decade, and Masterestaurant builds the system respecting that sequence. Add the return channel — 97% of SMS messages are read within 15 minutes of sending, according to Tabular (SMS Marketing Stats 2025) — and you will see that repeat business is triggered by a relevant message, not by free points.

4. Bands under 500 thousand and 500 thousand to 1 million

Below 500 thousand dollars a year, forget the points platform: your program is a phone list with consent and one message a month, because a 300-dollar monthly license eats 3,600 a year, close to 0.7% of revenue, without guaranteeing a single extra visit. The threshold here is simple: put no money into loyalty technology until 40 identified guests repeat at least twice per quarter. Operating target for this band: 25% of transactions identified with a name and a phone number, no discount involved.

5. Above 1 million: this is where incentive economics start

Past one million in annual revenue, the program stops being a list and becomes a P&L line that demands its own governance, with a hard ceiling of 2% of net sales allocated to redemption and a monthly review of incremental margin by cohort. Run the counterfactual all the way: if you give away a 9-dollar dessert with 2.70 in product cost to a guest who already came twice a month, you burn 64.80 dollars of margin per guest per year, and with 500 such guests that is 32,400 dollars walking out of the kitchen without moving a single additional visit. That same money, placed in owned content and a complete Google Business profile, performs differently: complete profiles are 7 times more likely to receive clicks, according to WebFX (2026). Cut the subsidy to your regular and buy back the dormant repeater.

6. Above 5 million and the celebrity-chef case

In the above-5-million band, including the media-chef restaurant or the large-format themed venue with 180 seats, the danger is not the discount but mistaking an image giveaway for a retention strategy, because the comped table for an influencer generates no cohort and does set a precedent. The decision here is architectural: segmented redemption by cohort rather than flat across the base, aiming to push transactional penetration to the 37% marked by the Paytronix 90th percentile (Loyalty Trends Report 2024). And one real tension in this band, resolved rather than dodged: a strong brand pulls new traffic that needs no incentive, while the program tends to reward exactly that traffic. You break it with eligibility rules — only guests who have not visited in more than 60 days qualify — and the typical saving lands near 40% of redemption spend.

7. Group or chain above 10 million

Above 10 million, loyalty is governed as a business unit with an owner, a budget and a separate accounting line, because the problem stops being the program and becomes consolidation across locations, channels and aggregators. Your own data is worth more than the commission you pay: in the United States, DoorDash closed 2024 with 60.7% of the delivery market, Uber Eats with 26.1% and Grubhub with 6.3%, according to Earnest Analytics, and every order routed through them is a guest you cannot contact again. With the global delivery market at 288.84 billion dollars in 2024 and a projection of 505.50 billion by 2030, per Grand View Research, migrating from third party to owned channel is a balance-sheet decision. Threshold for this band: 35% of digital sales through the owned channel within twelve months.

8. The decision that fits on one page

Before renewing the platform or signing the next loyalty contract, sit down with three numbers and decide in twenty minutes: real redemption cost for the last quarter expressed as a percentage of net sales, twelve-month frequency of the cohort that came twice and never returned, and percentage of identified transactions. If redemption exceeds 2% of net sales and the dormant cohort has not moved, the program is paying for what you already had. I defended the universal point for years as a cheap entry door, and I was wrong: content opens that door — 48% of operators were on TikTok in 2025 versus 26% in 2023, according to TouchBistro (State of Restaurants 2025) — while the incentive should only close a visit that was already hesitating. Start tomorrow by segmenting the base and switching off the regular's redemption.

9. Four differences that move the P&L

How the incentive is accounted for. A badly built points program is budgeted at the top and paid at the bottom: redemption hits the dish's contribution margin. The right method treats every point as a variable cost of sale and holds it to the same scrutiny as food cost, with 32% per dish as an absolute ceiling. What gets measured. Counting sign-ups says nothing; the metric is frequency by first-visit cohort and guest LTV over twelve months. With top-decile operators reaching 37% of transactions through loyalty (Paytronix, 2024), the benchmark that matters is transaction penetration, not database size.

10. Four differences that move the P&L — in practice

The order of the levers. The classic error is incentive first, content later, data never. The order that holds a sales funnel together is data, content, incentive, and in that sequence: 72% research on social before deciding (Restroworks, 2025), so content owns the top of the funnel and incentive merely closes one specific cohort. Who owns the channel. Handing the order to an aggregator hands over the relationship. With most consumers preferring to order direct, scalability depends on whose name the guest ends up under.