Repeat-purchase program: the traditional method against the local digital engine at Masterestaurant

1. What decides whether a repeat-purchase program returns margin or burns it?

The channel the guest comes back through decides it, not the reward you hand over. With half of a restaurant's revenue typically depending on people who already know the house, that half arrives by routes with very different costs:

a guest ordering on your own online channel buys more items per check than one coming through a third-party app, so rewarding both visits equally gives away margin on the expensive one. The boardroom mistake I see repeated is arguing about the discount percentage before knowing which channel brought the previous visit. Order the MEASUREMENT by channel first, the incentive second. A program without that traceability is not a program: it is a promotion with a card attached.

2. Under 500 thousand a year: no points, one list and one channel

In the band under 500 thousand dollars a year the decision is to skip the points program and build a LIST. The operating threshold is plain: if your permission-based contact base does not reach 800 names, any loyalty platform charges more than it returns, because the redeemer will be the guest who was coming anyway. Collect phone and email at the register and on your own ordering channel, then work the send. SMS opens a lot and gets read fast; restaurant email opens reasonably but converts to clicks very poorly. At small volumes SMS wins outright and demands no new software. Rule: two sends a month, none of them a flat discount.

3. From 500 thousand to 1 million: the local engine before the punch card

Between 500 thousand and 1 million, the investment that pays back first is the local profile, not the stamp system. Once a guest has seen a business's page, the window to contact or visit it is a matter of hours, not a brand campaign. Update the listing, fresh photos every month, an answer to every review inside 48 hours, and only then measure how many returning guests came from there. My threshold for moving to formal loyalty is 2.500 identified guests and 25% of sales through owned channels. Below that, a points program adds friction at the register and accounting noise without moving frequency at all.

4. The real cost of 20% off a dish carrying 30% food cost

A 20% discount does not cost 20%: it takes almost half the dish's contribution margin. At 30% food cost, every dollar sold leaves 0,70 of gross margin; cut 0,20 and margin drops to 0,50, meaning you gave up 28,6% of that dish's margin to win a visit you probably already had at 55% retention (Restroworks, 2025). That is why a non-monetary incentive —table priority, an off-menu plate, the nine o'clock Saturday reservation— costs operations rather than margin, and holds price. Here I was wrong for years, recommending two-for-one as the Tuesday hook. It fills seats beautifully and it wrecks the weekend price reference, which is exactly where the house profit lives.

5. Above 1 million: channel cohorts and a capped repeat-purchase budget

Past 1 million, the house can afford analytics and must budget repeat purchase as a line item, capped at 3% of annual sales. The decision architecture changes: instead of guessing that people want discounts, you measure cohorts by month of first visit and answer with channel, date and check when frequency drops in August. It helps to know 29% of United States restaurant traffic already arrives with some kind of offer (Circana, 2025) and that value menus grew 1% while total traffic fell 1% (Circana, 2025): discounting is the market floor, not an advantage. The advantage sits in migrating third-party orders to owned channels, where the check rises. Set the goal in mix points, never in redemptions.

6. Above 5 million: the media-driven profile and the borrowed-traffic trap

In the band above 5 million you meet the high-end case —the celebrity-chef room, the large-format themed venue— and its problem runs the other way: discovery is abundant, repeat purchase is scarce. Some 51% of TikTok users dine out because of a restaurant's content (Restroworks, 2025) and 38% of Gen Z discovery happens on that network (Toast, 2026, survey of 1.466 adults), so the first visit arrives on its own. The second does not. Diego F. Parra, consultant at Masterestaurant, pushes these houses toward one indicator and one only: share of guests with two visits inside 90 days. If that number lives below 12%, the house owns no brand, it rents traffic, and the day the algorithm shifts the evening turn empties out.

7. Group or chain above 10 million: a single customer identity

Beyond 10 million, the decision stops being which reward to give and becomes where the guest lives, and the answer is one identity across locations, owned delivery and reservations. Without it, the same diner appears three times and you pay acquisition three times over. The instrument with the cleanest accounting is the gift card, which carries a breakage rate near 6% of unredeemed value (Capital One Shopping, 2026): cash flow pulled forward and a slice that never leaves the kitchen. Budget SMS for transactional messages, where response reaches 45% against email's 6% (Omnisend, 2025), and leave email for content. One number that opens a door: measure what it costs to bring back a known guest, then compare it against what generic paid ads typically return in this sector.

8. What to do Monday morning, with a numeric threshold

Start by counting how many guests you can identify today by name, phone and last visit date, then divide that number by your monthly checks. If the result falls under 20%, you do not have a program problem, you have a capture problem, and no software fixes it. The sequence that works is short: identity at the register, local listing current because online reputation drives visits in a short window, owned ordering pushed by those extra items per check, and the incentive last of all. Invert the order and you will pay rewards to strangers. My recommendation is firm and has no middle ground: switch off any flat discount running this month, replace it with access, and measure 90-day frequency before touching price again.

9. What really separates the two models?

The first difference is decision architecture:

the traditional model decides on a hunch («people want a discount») while the local digital engine decides on cohorts, so when the owner asks why frequency dropped in August there is an answer with channel, date and check size, not a shrug. That answer is what lets you renegotiate rent, staffing and purchasing, because cash flow stops being a monthly mystery. The second is unit economics. A 20% discount on a plate carrying 30% food cost does not cost 20%: it eats close to half that plate's contribution margin, and if the loyal guest redeems it —the one who was coming back regardless— the house is paying for a visit it already had.

10. What really separates the two models — in practice

Non-monetary incentives cost operations, not margin, which is why the Masterestaurant method exhausts convenience before touching price. Third comes the channel. Restaurant email opens reasonably well yet converts poorly, while SMS answers at 45%, according to Omnisend (2025). When you have a soft Tuesday and need thirty covers tonight, the gap between those two numbers is literally the gap between closing that shift in the black or in the red. Fourth is territory risk. The traditional engine cannot tell a guest living six blocks away from someone passing through; the digital one can, and that is where geotargeted media and the delivery algorithm start working together, because a tightly drawn delivery radius lifts on-time rate and on-time rate is exactly what Rappi or Uber Eats reward with visibility. Smaller radius, more repeat purchase.