Five years of live data: what repeat visits really cost
The patterns hide in plain sight — weekday lunches, festive peaks, the perfectly-timed reminder. Years of real production data, read honestly, change what you optimise for.
Most operators can tell you their best Friday of the year. Far fewer can tell you how many of last Tuesday's lunch guests came back. That gap — between the nights you remember and the visits you can't see — is where the real economics of a restaurant live, and it only becomes visible once you've watched the same guests, in the same venues, across years rather than weeks.
We've had that vantage point. Years of live production data across NJ Group's own restaurants and hotels — not a survey, not a benchmark deck, but the actual orders, reservations and reminders running through real services. Read in aggregate, the patterns are unglamorous and a little humbling: the visit that matters most is rarely the one you celebrated, and the cheapest growth you have was sitting in your own data the whole time.
- Acquisition is a cost you pay again and again; a repeat visit is margin you've already earned.
- Weekday lunch — not the festive peak — is where retention is quietly won or lost.
- Timing beats offers: the same reminder is worth a multiple of itself when it lands on cadence.
- You can only act on a pattern you can see — and you only see it if you own the data.
01 — The visit you don't celebrateRetention is cheaper than acquisition — by an order you can feel
Every new guest arrives with a price tag — an ad, a discount, an aggregator's cut of 15–30% to be discovered at all. That cost is real and it recurs: win a guest once and, on a platform, you often have to win them again the same way next month. Across the data, the venues that grew steadily weren't the ones acquiring fastest. They were the ones losing guests slowest.
A returning guest skips that entire toll. No introduction, no incentive to overcome unfamiliarity, no rented reach — just a person who already chose you, choosing you again. Stack a year of those visits and the maths is plain: a small lift in how often your existing guests come back outperforms a large lift in how many new ones you find. The repeat visit is the one you should obsess over, precisely because it's the one nobody throws a party for.
02 — Where the patterns actually liveThree signals the production data kept surfacing
Read across enough services and the noise falls away. The same three signals showed up again and again — none of them flashy, all of them actionable, and every one of them invisible to an operator who only owns a payout statement.
- 01
Weekday lunch is the loyalty engine
Festive peaks fill the room once; the weekday-lunch regular fills it fifty times a year. The strongest repeat behaviour clustered in the unremarkable midday slot — the habit, not the occasion.
- 02
Cadence has a half-life
A guest's likelihood of returning decays on a clock you can almost set a watch to. Reach them inside that window and they come back; let it lapse and you're paying to acquire them all over again.
- 03
The right nudge beats the big offer
A timely, relevant reminder consistently outperformed a deeper discount sent at the wrong moment. Margin wasn't lost to generosity — it was lost to bad timing.
"The festive spikes are loud. The steady swell beneath them is what actually pays the rent."
03 — The quiet art of the timed reminderWhy timing is a discipline, not a campaign
The temptation, once you see a guest go quiet, is to shout — a blast to everyone, a blunt discount, a hope that volume covers for precision. The data argued the opposite. The reminders that worked were narrow: the right guest, near the end of their natural cadence, with a reason that fit what they actually order. Sent to everyone at once, the same message trained guests to ignore you and discounted the visits you'd have got for free.
Doing that by hand across a full guest book is impossible, which is the real reason it rarely gets done. So it has to be governed by a system that acts within rules you set — when to reach out, to whom, with what, and never below a margin you'd defend. We call that Policy-Driven Intelligence: not a tool that blasts on a schedule, but a decision engine that respects each guest's rhythm and your guardrails at the same time.
That's the difference between a campaign and a discipline. A campaign is a thing you launch; a discipline is a thing that runs. The venues that turned these patterns into repeat visits weren't sending more messages — they were sending fewer, better-timed ones, continuously, without a person having to babysit the calendar.
04 — You can't act on what you can't seeThe patterns are worthless if the data isn't yours
None of this is visible from a marketplace dashboard. Aggregators hand you a payout and a rating; they keep the guest, the cadence and the context. You can sense that lunch is busy, but you can't see that a specific cohort of regulars has started slipping — and by the time it shows up in revenue, the pattern has already cost you. Owning the data is what turns a vague worry into a visible, fixable signal.
This is why local context isn't decoration. Singapore dines on its own clock — payday peaks, public-holiday swings, the hawker-to-fine-dining spread — and a model that ignores those rhythms reads every quiet Tuesday as failure rather than season. Years of live production data taught the system our rhythms first, because the first venues it had to get right were our own.
Five years in, the lesson is almost dull in how plainly it repeats: the cheapest, most durable growth a restaurant has is the guest it already earned, coming back on time. Acquisition makes the noise. Retention pays the bills. The operators who internalise that — and own the data to act on it — stop renting their growth and start compounding it. That's what it means to own your table.