The standard advice is to reply to bad reviews quickly. It is good advice and it quietly skips the step that determines whether the reply is any good.
A review arrives describing a disappointing Saturday. You have two options. You can write the reply everyone writes — so sorry to hear this, that's not the standard we aim for, please do get in touch — which is fine, generic, and commits to nothing because you know nothing. Or you can spend eight minutes finding out which table it was, and write something that demonstrates you did.
The second reply is not just warmer. It is read by prospective guests as evidence that the restaurant knows what happens inside it, which is the actual thing they are trying to assess.
What a review gives you to work with
More than it looks. Reviews leak identifying detail constantly, and five minutes with the right records narrows most of them to a single table.
| The review mentions | Look it up in | What you learn |
|---|---|---|
| A day and rough time | Booking system, cover report | Which parties were in that window |
| A party size | Bookings | Usually narrows it to two or three tables |
| A named dish | Tickets for that service | Frequently narrows it to one table |
| An occasion — birthday, anniversary | Booking notes | Almost always identifies it outright |
| A dietary requirement | Booking notes, allergen log | Identifies it outright |
| A wait of a stated length | Ticket times | Confirms or contradicts, and tells you the section |
| "The tall chap with the beard" | Rota | Who was on, and which section |
| A card payment, a refund, a comp | Till and void records | The exact transaction |
An occasion or a dietary note is close to a unique identifier. A dish plus a time is usually enough. Even a review with nothing but "Saturday night" narrows to a service, and the service-level records are often sufficient — if ticket times that night were double normal, you know what happened without finding the table.
The eight-minute sequence
1. Establish the service. Which date, which sitting. If the review does not say, use the posting date and work backwards: most reviews are written one to three days after the visit, so the preceding weekend is the first place to look.
2. Pull the service records. Covers, ticket times, the rota as worked, voids and comps, and whatever the duty manager wrote in the log. Before looking for a table, find out whether the whole service was unusual. A night with ticket times 40% above normal explains a review on its own and changes what you write.
3. Narrow to a table. Party size and time, then the dish, then the occasion. Three filters resolve most of them.
4. Ask whoever was on that section. The most valuable step and the one that gets skipped, because it requires finding a person rather than opening a report. Servers remember bad tables with surprising precision, and they frequently know something no record holds: the party arrived forty minutes late and expected the same pace, the kitchen lost the ticket, a child was unwell, someone was already angry on arrival.
5. Establish whether it was addressed at the time. Was anything comped, was a manager called, was anything said. This single fact changes the reply more than any other, because a guest who complained in the room and was fobbed off is a different situation from a guest who said nothing and wrote it down later — and only one of those is a service-recovery failure.
6. Decide what you can say. Some of what you now know cannot be used publicly, and the line matters.
What you can and cannot say in public
This is where identification can go wrong, and the rules are firm.
Do not reveal that you identified them. "I can see you booked for your wife's birthday and sat on table 14" is accurate, reads as surveillance, and will cost you more than the original review. The guest did not consent to being matched to a booking in public.
Do not name staff. Not to defend them and certainly not to blame them. "Dani is normally excellent" puts a named employee into a public dispute.
Do not use the detail to argue. The temptation after a good investigation is to correct the record — our records show you were seated at 19:40, not 19:10. Never. You will be right and it will read appallingly, to everyone except you.
Do use it to be specific about the cause and the change. This is the entire payoff:
You're right that Saturday ran badly — we were short on the pass from about half seven and food was sitting too long before it got to tables. That's on us, not on the kitchen being busy. We've changed the running rota for Fridays and Saturdays from this week. If you'd give us another go, ask for Sam and I'll look after you.
Compare that with the generic version. It concedes something specific, it attributes the cause to a system rather than a person, it names a change, and it is plainly written by somebody who looked. None of it discloses that the booking was identified.
When you cannot identify it
Three honest cases.
No usable detail. A one-star with no text, or "terrible, never again". Nothing to investigate. Reply briefly, invite contact, move on.
Nothing matches. No booking, no ticket, no transaction, and the dish was not on the menu that week. Possibly a misremembered date — check the surrounding fortnight before concluding anything, because this is the most common false positive. If genuinely nothing matches, that is a documentation exercise rather than a reply, and the reply should stay neutral while you build the file.
It is a courier or a platform problem. Off-premise reviews frequently describe something that happened after the bag left. Identify the order, confirm what was packed, and then reply acknowledging the experience without accepting a cause you have ruled out.
Why this beats speed
Replying within the hour with nothing is worth less than replying the next morning with something. The reply is permanent, public, and read mostly by people deciding whether to book — and what they are assessing is not your response time. It is whether this restaurant seems to know what goes on inside it.
There is one exception, and it is categorical: anything alleging physical harm — an allergic reaction, a foreign object, illness after a meal — is not a reply problem and does not wait for an eight-minute investigation. That has its own first-hour sequence and it starts immediately.
The ways identification goes wrong
It depends entirely on your records. If covers are not timestamped, the rota lives in a photo, and nobody writes a duty log, step two is unavailable. That is worth fixing for reasons well beyond review replies.
Eight minutes becomes forty. Identification is interesting, and the pull towards reconstructing an entire evening is real. Set a limit: if three filters have not narrowed it, reply generically and stop.
Not every review deserves it. Reserve this for reviews describing something specific and checkable, and for anything you suspect is a pattern. A mild three-star about parking does not need a ticket lookup. Deciding what gets answered now, later or never is the prior step.
Asking the server can go badly. "Do you remember table 14 on Saturday" is a reasonable question. Asking it in front of the team, or in a tone that implies the review is their fault, converts an investigation into a telling-off and guarantees worse information next time.
The staff account is not necessarily right either. You now have two partial versions. The point is not to establish who was telling the truth — it is to find the operational cause, which is usually visible in both accounts and is almost never a person.
OMMU attaches the service's covers, ticket times and staffing to each review as it arrives, which collapses most of steps one to three. Without it, the records exist anyway — the discipline is simply opening them before opening the reply box.