The restaurant repeat-purchase program: before and after, with the numbers on the table

Step 1: build the identified-customer base with each last-purchase date

Your first deliverable is not a campaign but a table: name, phone, source channel and last-purchase date for every customer, and you know it is finished when you can filter in thirty seconds who has not returned in more than forty-five days. That table belongs to you, unlike what happens inside delivery apps, which charge between 18% and 30% commission and never hand over the diner's phone number. Start with what already exists: the reservation log, WhatsApp orders, the till entries that carry a name. Fifteen hundred identified customers in a casual-service location is enough of an asset to work with, and adding fifty to seventy contacts a week through counter capture is perfectly realistic. Global online food delivery is projected at US$1.51 trillion for 2026 according to Statista Market Forecast 2026, and you take no part in that volume if you cannot say who bought from you yesterday.

Step 2: segment by frequency and set a churn threshold with numbers, not intuition

Segmenting by frequency means splitting the base into four buckets — active, cooling, at risk and dormant — using the real gap between visit one and visit two in your own location, which in quick and casual service usually lands between twenty and thirty-five days. The measurable deliverable is a written threshold: if your house average is twenty-eight days, a customer turns at-risk on day forty-two and dormant at ninety. Somewhere between 60% and 70% of one-time buyers never come back within that quarter, and the leak shows up in no report at all, because the point of sale counts tickets rather than people. I got this wrong for years, treating repeat purchase as a Tuesday-discount matter that ate margin without moving a single visit. Verify the step by counting how many customers sit in each bucket: if 80% are dormant, your problem is not ad spend.

Step 3: build the audiovisual content calendar that creates the trigger

Desire triggers repeat purchase, not the coupon, and you manufacture desire with a calendar of three to five weekly pieces published in the decision windows: eleven in the morning for lunch, six in the evening for dinner. A thirty-second Reel showing a brisket being sliced, sauce dropping, the knife audible, moves orders without touching food cost, which on your menu should never run past 32%. Evidence backs this up: 58% of consumers visited a restaurant after seeing it on TikTok, against 38% in 2022, per the MGH Survey 2024. And brands with the strongest social strategy reported 14.1% higher revenue according to Deloitte Digital. Your deliverable is the calendar loaded with twenty pieces for the month plus their time slots; it is verified once you publish four straight weeks without a single gap. The offer fires on behavior, never on the calendar, and that distinction separates a repeat-purchase program from a WhatsApp list that annoys people.

Step 4: fire the value offer only when the customer crosses the risk threshold

When a customer crosses day forty-two without returning, out goes a personal message with their name and a concrete reason — the new dish, grill night, the table they liked — and only if fifteen days later they still have not shown up does the economic incentive enter, capped and with an expiry date. Channel matters: 97% of SMS messages are read within fifteen minutes of sending, and a reservation confirmation message generates US$4.20 in revenue, according to Tabular SMS Marketing Stats 2025. Waking a dormant customer costs you the message; bringing in a stranger costs between US$9 and US$22 per new customer. Your deliverable: the rule written down and automated, with a weekly count of sends and replies. A properly built program lifts frequency by 0.4 to 0.9 visits per customer per quarter and cuts acquisition cost by 25% to 40%, though those numbers mean nothing unless you measured the baseline first.

Step 5: measure frequency and acquisition cost before and after, using the same cut

Take the previous quarter, divide total visits by unique identified customers, and store that ratio somewhere it will not get lost. Then measure the same ratio ninety days later, under the same identification criteria, because switching the denominator midway is optimism's favorite trick. Some 75% of quick-service brands running loyalty programs reported higher traffic in 2025 according to the National Restaurant Association, and with sector net margin that Statista places between 3% and 9%, half an extra visit per customer decides your year. The step is verified when both ratios sit on one sheet. Training your customer on discount is the costliest error: anyone who receives a promotion every fifteen days learns to buy only on promotion, and you end up selling the same volume on thinner margin. Second comes contact frequency without criteria, blasting the whole base every Friday until people mute the number; talk to active customers through content and save the direct message for those who crossed the threshold.

The mistakes that sink the program, and how to get out of each one

Third is failing to ask permission and never cleaning the base, leaving dead phone numbers that inflate reach and hide the real decline. Fourth, and the one that surfaces most in audits, is handing repeat purchase to the delivery app, which keeps the data and 18% to 30% of every ticket. Given that 92% of diners read reviews before choosing where to eat, per Restroworks, your own base competes against that noise with one advantage: they have already tasted your food. Picture reactivation working and Tuesday no longer being a dead day: four hundred dormant customers return within six weeks, a kitchen sized for eighty covers takes on a hundred and twenty, ticket times stretch to thirty-five minutes, and the reactivated customer — arriving with high expectations precisely because you wrote to them personally — walks away with the worst experience they have ever had with your house. That customer will not fall for a second message.

What would happen if the program works and your kitchen was not ready?

This is why the MASTERESTAURANT method I use with owners orders capacity ahead of demand:

stagger reactivation in batches of one hundred to one hundred fifty contacts per week, measure ticket time in every batch, and open the next batch only if service held. Growing demand beyond your operation is not growth, it is manufacturing detractors out of your own database. You know the program is built when five questions get answered without opening two systems: how many identified customers you hold, how many days pass on average between visit one and visit two, how many crossed the risk threshold this week, how many content pieces went out in the last thirty days, and what it costs today to bring in a new customer compared with waking one of your own. If all five answers live on a single sheet updated each Friday, the system exists; if any of them depends on the manager's memory, it does not yet.

Closing checklist: how to know everything landed properly

Diego F. Parra and the Masterestaurant team review that sheet before the income statement, because frequency is a leading indicator and margin is a lagging one. Begin with the only thing that matters this week: sit down today with the last ninety days of records and calculate your gap between visit one and visit two. Difference one is data ownership: with scattered promotions the guest belongs to the delivery app, which charges 18% to 30% commission and never hands over the phone number, while a repeat purchase program builds a first-party base where the marginal cost of reaching ten thousand people approaches zero. Whoever owns the base sets the price of their own demand. The trigger is difference two. A promotion pushes with a discount, and discounts train guests to wait for discounts; short-form video pushes with desire, and desire never erodes margin. Thirty seconds of a brisket being sliced at eleven in the morning drives more lunch orders than a 20% coupon, and it leaves food cost untouched, which on your menu should never cross 32% per dish.

Five differences that move the cash

Third, recency segmentation rewrites the message itself. Guests inside the first fifteen days hear about what is new, the 46-day crowd hears about what they are missing, and anyone past ninety days gets a win-back offer with a short expiry. Sending one identical message to all four segments is the fastest route to being muted. Fourth difference, and almost nobody executes it: review timing. Asking within ninety minutes of the meal triples response rate against the generic next-day request, and those reviews feed the online reputation that later decides delivery conversion, since platform ranking weights recent ratings. The fifth is governance. A real program has a named owner, a twenty-minute weekly meeting and three numbers on the wall: average frequency, share of identified tickets, and sales attributed to the first-party base. Without those three visible, the program degrades into one more broadcast list and dies in month two.