Every review comparison rests on a division that almost nobody performs. Without it, the answer to "which site is worst" and "which shift is worst" is always the biggest one, and it is always wrong.
The division is complaints divided by covers. It takes ten minutes to set up and it changes most of the conclusions people draw from their reviews.
The error, stated plainly
Site A: 300 covers a week, 6 complaints. Site B: 90 covers a week, 4 complaints.
Raw counts say A is the problem, by half again. Per hundred covers, A runs at 2.0 and B at 4.4 — B is more than twice as bad, and it is the site nobody is discussing, because it produces fewer angry emails.
The same arithmetic applies to dayparts within one site, and this is where it bites hardest. Friday evening generates the most complaints in essentially every venue on earth, and Friday evening is also 25 per cent of the week's covers. Reacting to that without dividing is how operators spend money on their busiest shift while a genuine fault sits undisturbed on Sunday lunch — an actual case we have written up in the Friday pattern that was not real.
Reviews per cover is very low, and that matters
Typical review rates run somewhere around one review per 30 to 100 covers, and negative reviews are a fraction of that. So the numbers you are dividing are small, and small numbers are volatile.
Two consequences worth taking seriously:
Do not read a single week for a single daypart. A shift with 90 covers might produce one complaint or three by chance. The month is the smallest honest unit for daypart comparison, and the quarter is better.
Quiet shifts need longer windows than busy ones. Sunday lunch at 60 covers needs about three months to say anything; Friday evening at 300 needs one. This is unintuitive and it is the reason problems survive on quiet shifts: they are statistically harder to see, so they go unexamined for years.
If you want a rough rule, wait until the denominator is above roughly 500 covers before you take a rate seriously.
Setting it up
You need two tables and neither is difficult.
Covers by site by daypart. From the till. Once, then updated monthly. Most operators have never assembled this in one place, and it is useful for a dozen things beyond reviews.
Complaints by site by daypart. From the reviews, tagged as they arrive. The daypart usually has to be inferred — guests mention the day more often than the time, and party size and menu references help.
Then a rate per hundred covers for each cell. A group of six sites with four dayparts is a 24-cell table, which sounds like a lot and fits on half a page.
Reading it
You are looking for three things, in this order:
Cells well above the group or venue average. Two times the average is worth investigating. Ten per cent above is nothing.
Cells that have moved. A site going from 1.2 to 2.6 is more urgent than one sitting steadily at 3.0, because recent changes have findable causes. The steady 3.0 is a structural problem you have been living with; the move happened for a reason someone can remember.
Cells with almost no complaints. Genuinely useful and always skipped. A daypart at 0.3 per hundred covers is either doing something right that could be copied, or is not collecting reviews — and those two possibilities want very different responses. Check the review count before you celebrate.
Where the division misleads
Two honest limitations.
Covers is not always the right denominator. For anything about waiting, turnover or noise, the relevant driver is occupancy at a moment rather than covers over a week. A site at 70 per cent capacity all evening behaves differently from one that is empty at six and full at eight with the same total. Where you can, use covers in the specific hour.
Party size shifts what gets complained about. A site with lots of large tables generates more handoff and timing complaints structurally — see the cold food case study — and that is a real operational difference rather than a fault. Comparing a group-booking site to a couples site on timing complaints will mislead you in a way the division does not fix.
What it is for
Not to produce a league table. The point is to stop the loudest shift consuming every conversation, which it will otherwise do indefinitely, because reviews arrive with emotion attached and the Friday ones are the ones you remember.
Once the rate exists, the arguments get shorter. "Sunday lunch is running at nearly twice our weekend rate" ends a discussion that raw counts can sustain for months.
Keeping it current across sites and months is the part that does not survive a trading year, and it is precisely the part that makes the numbers worth anything — a rate with no history cannot tell you what moved. That continuity is what OMMU maintains in the background, so the comparison is available in the week you need it rather than reconstructed from scratch each time.