Most restaurant groups file reviews under marketing, and the reasoning is sound. Reviews are public, they are read by prospective guests, they look like communications, and marketing is the function that handles things the public reads.
The consequence is also predictable, and it is not that anyone does their job badly. It is that the organisation ends up extremely good at responding to problems and structurally incapable of removing them.
This is a question of ownership and incentives rather than effort, and it is worth separating from the tactical question of who physically writes the replies — which is a different and also neglected problem.
What a function is measured on determines what it can see
Marketing is accountable for how the business appears. Given a review saying the chicken was dry, the available levers are: reply well, reply quickly, and over time improve the ratio of good reviews to bad ones.
None of those levers touch the chicken.
Not because the marketer does not care — they will often flag it, sometimes repeatedly. But flagging is the end of their authority. They cannot retime a spec, change a supplier, move a start time, or replace a grill. So the review gets a thoughtful reply, the issue gets mentioned in a weekly summary, and the next one arrives.
Operations has the opposite problem. It holds every lever and does not see the data, because the data arrives in a channel owned by somebody else and is summarised into a slide by the time it gets near a kitchen.
So the information sits with the function that cannot act, and the authority sits with the function that cannot see. Everything downstream follows from that single split.
The symptoms, which are recognisable
If this is your arrangement, you will have some of these:
Response coverage is excellent and complaint themes are stable. The clearest signature. 90% coverage, replies within hours, well written — and the same four themes for two years. The process is working perfectly at the thing it is actually set up to do.
The reply is reported as the resolution. "Handled" means a reply was posted. There is no field in the workflow for what changed in the restaurant, so nothing is recorded there, so nothing is expected there.
Chefs and GMs have never read a review in full. They get a theme count and a sentence. The original text — which is where the useful specificity lives, because a guest saying the chicken was dry the second time, the first was fine has told you something a theme count destroys — never reaches the only person who could act on it.
Review data is absent from operational meetings. It appears in a monthly marketing report, alongside social engagement and campaign performance, which is the correct home for a channel metric and the wrong home for a defect report.
Nobody can name a change caused by a review. Worth asking directly. If the honest answer across a year is nothing, the function is cosmetic — not through incompetence, but by design.
What moving it changes
The proposal is not to take replies away from the people who are good at writing them. Marketing is usually better at the public wording and should keep it. What moves is accountability for the underlying cause.
Concretely, three changes:
The theme report goes to operations with the review text attached. Not a count. The actual sentences, grouped. A head chef reading eleven reviews about dry chicken does something different from a head chef told that "food quality mentions are up 4%", because the eleven sentences contain the pattern — the dish, the daypart, the second visit — and the percentage does not.
Every theme above a threshold gets an operational owner and a date. A name, a change, a date. Not an action plan, which is where this degenerates; one change, one person, one week.
The success measure becomes theme rate, not response rate. This is the pivot the whole thing turns on. As long as the number being reported upwards is response coverage, a function optimising rationally will produce more and faster replies. Make the reported number the complaint rate per hundred covers and the same rational optimisation points at the kitchen.
The objection, which is fair
Replies matter. They do. A good reply is read by prospective guests, affects booking decisions, and sometimes recovers a guest who would otherwise not return. Replying well is a real skill worth having, and answering everything genuinely does more for you than chasing a higher star average — we have argued that at length.
The argument is not that replies are worthless. It is that replies are the only thing a marketing-owned process can produce, and a business that only does the thing its structure permits will mistake that thing for the whole job.
The honest formulation: replies are how you manage the review you already have. Operational ownership is how you receive fewer of them. You want both, and only one of them is a function of who writes the reply.
The version that works in a single site
All of the above is organisational and assumes separate functions. In an independent restaurant the owner is marketing, operations and often the chef, so there is no handover to get wrong.
The equivalent failure is temporal rather than departmental. Reviews get read on a phone, in the ten minutes before service, in reply mode — the only mode available at that moment. The reply gets written, the pattern never gets assembled, and the same complaint is answered gracefully fourteen times.
The fix is a different slot rather than a different person. Twenty minutes, weekly, not adjacent to a service, reading the week's reviews as a set rather than as arrivals. Four lines is enough to write down what it produces. The constraint being removed is the same one: you cannot do pattern-finding in the posture of answering.
The honest limits of this argument
Operations will not want it. Reviews arrive as criticism, often of named people, in public. A function handed this without the authority and time to act on it will treat it as an additional reporting burden, correctly. Move the data and the authority together or do not move it.
Theme rate is slower and less flattering than response rate. Response coverage can be improved in a fortnight. Theme rate moves over quarters, and for the first few months of honest measurement it will not look like progress. Anyone who needs a monthly win will prefer the old metric.
Some reviews genuinely are reputation problems. A pile-on, a coordinated attack, something that has reached local press — those are communications events and belong with the people who handle communications. Telling a burst apart from a bad week is its own skill, and getting it wrong in the other direction is also expensive.
Nothing here is about anyone's competence. Marketers who own reviews usually know exactly what the operational causes are and have said so. The structure is what prevents the sentence turning into a change, and reorganising the structure is the only version of this that works.
The question worth asking at your next operations meeting is simply: name a change we made in the last year because of a review. Where there is a good answer, the ownership is right regardless of the org chart. Where there is a long pause, the pause is the finding.
OMMU delivers themes with the underlying review text to the person who can change the thing, which is mostly a routing problem rather than an analytics one. A weekly email with the sentences in it does the same job, as long as it goes to the kitchen rather than to a report.