Most case studies end with the fix working. This one ends with an operator six weeks and roughly £4,000 of extra labour down, having solved a problem that was never there. It is more useful than the success stories, because the reasoning that led them wrong is the same reasoning everyone uses.
Composite, with representative numbers.
What the data appeared to say
Over six weeks the venue collected 34 negative reviews. Nineteen of them — 56 per cent — described a Friday or Saturday evening visit. The complaints were about pace: slow to be greeted, slow to order, long gaps.
The operator did the sensible thing and added a floor host and an extra runner to both weekend evenings. Cost, about £330 a week. The expectation was that weekend negatives would drop toward the weekday rate within a month.
They did not move at all. Six weeks later weekend reviews were still 54 per cent of negatives.
The two mistakes, both of them ordinary
They compared complaint share to nothing. Friday and Saturday evening were 48 per cent of this venue's covers. So 56 per cent of complaints from 48 per cent of covers is very close to what you would expect if service quality were identical across the week. The apparent pattern was almost entirely a volume effect. Adjusted per hundred covers, weekend evenings ran at 1.9 negatives and the rest of the week at 1.6 — a real but small gap, nowhere near large enough to justify the intervention.
This is the single most common analytical error in reading reviews, and it is invisible unless you deliberately do the division. Your busiest shift will always generate the most complaints. It is your busiest shift.
They read "slow" as one complaint type. Sorted properly, the weekend comments split into two groups that want opposite fixes. Eleven described waiting to be seated or greeted — a door and host problem. Eight described long gaps between courses once seated — a kitchen throughput problem at peak. The extra runner helped neither, because runners move plates that already exist and the kitchen was the constraint.
Where the pattern was genuinely strong, and nobody looked, was Sunday lunch: 6 negatives from 9 per cent of covers, which works out at 2.9 per hundred covers. Nearly twice the weekend rate. Sunday lunch ran on the thinnest rota of the week and had one long-standing problem with roast timings that everybody on the floor knew about and nobody had escalated, because Sunday only ever produced two or three bad reviews a month and two or three feels like noise.
The real finding was sitting in the smallest bucket. Raw counts had hidden it and the weekend was loud enough to absorb all the attention.
The three questions that would have stopped this
Before acting on any pattern in reviews, in order:
- What share of covers does this shift represent? If the complaint share is not meaningfully higher than the cover share, you do not have a pattern. This one question would have prevented the whole episode.
- Do the complaints describe the same mechanism, or just the same mood? "Slow" is a mood. "Waited twenty minutes at the door with empty tables visible" is a mechanism. Only mechanisms are actionable, and a cluster of moods usually means several different faults, each too small to see.
- What is the smallest bucket, per hundred covers? Quiet shifts produce few reviews and are therefore systematically under-investigated, which is exactly why problems survive there for years.
The Friday case we wrote up separately passed all three: 63 per cent of complaints from 19 per cent of covers, a repeated and unusual mechanism, and a mechanism that matched the rota. The difference between that case and this one is not the quality of the operator. It is whether the division got done.
What the six weeks was actually worth
Not nothing. The host improved the door experience and stayed. The exercise produced a covers-by-daypart figure the venue had never assembled before, which is what made the Sunday finding visible afterwards. And it established that weekend pace was, at worst, marginally worse than the rest of the week — which is a legitimate thing to know.
But the honest accounting is that a plausible story got funded because the raw counts agreed with what the operator already suspected. Reviews are very good at confirming a theory you brought with you. They arrive one at a time, with emotion attached, and the Friday ones stick in the memory in a way the Sunday ones do not.
The habit worth building instead
Keep one small table — covers by daypart — and never look at a complaint count without it. That is genuinely the whole discipline. Complaints per hundred covers, by shift, month over month. It takes the drama out of the reading and it stops the loudest shift monopolising every conversation.
It also has to happen every month to be worth anything, because a single month's figures on a quiet shift are far too small to read. That continuity is the part that fails in practice, and it is what OMMU is for: the normalising is done in the background, so what lands is "Sunday lunch is running at nearly twice your weekend rate per cover" rather than fifty reviews and a hunch.