Eleven sites. One of them, Brierfield, was producing dry-chicken complaints at nine times the group rate — 2.7 per hundred covers against a group average of 0.3.
Every available instinct says Brierfield. The GM had been in post five months, the site had the oldest kitchen in the estate, and it had been on the watch list since spring. The area manager had a plan, and the plan was reasonable: two days on site, retrain the grill section, replace the probe thermometers, revisit the GM's development plan.
What stopped it was one check that took about fifteen minutes, and the finding reversed the whole quarter.
The check
Before driving anywhere, they asked a different question: not "why is Brierfield worse", but "is this theme rising anywhere else".
Dry-chicken and overcooked-chicken mentions, per hundred covers, by month, across all eleven:
| Site | May | Jun | Jul | Aug |
|---|---|---|---|---|
| Brierfield | 0.4 | 0.9 | 1.8 | 2.7 |
| Shawcross | 0.1 | 0.2 | 0.5 | 0.8 |
| Thornleigh | 0.0 | 0.1 | 0.4 | 0.7 |
| Granby St | 0.2 | 0.2 | 0.3 | 0.6 |
| Other seven | 0.1 | 0.1 | 0.2 | 0.4 |
| Group | 0.1 | 0.2 | 0.3 | 0.6 |
Brierfield is the outlier and that is not in dispute. But look along the bottom row. The group rate had sextupled in four months. Every site was climbing. Brierfield was climbing fastest and from the highest base, which is what made it visible — and which had drawn all the attention to the one place the cause definitely was not.
A group-wide rise does not come from eleven kitchens independently getting worse at the same thing in the same quarter. It comes from an input.
What the input was
Procurement had changed chicken supplier on the first of May.
The new breast was leaner and about 15% lighter. Same price per kilo, a slightly better margin, and on paper an equivalent product. It had passed a tasting in March. What nobody had done was retime the cook — the spec still said six minutes a side, which had been right for the old, heavier, fattier breast and was comfortably too long for the new one.
Every site in the estate was overcooking chicken, by the book, correctly following a spec that had become wrong on 1 May.
Why Brierfield was nine times worse at the same fault
This is the part worth keeping, because it generalises far beyond chicken.
Brierfield was not making a different mistake. It was making the same mistake with the least margin for absorbing it:
- The oldest grill in the estate, running hotter at the back than the front. Every other site had some of this; Brierfield had the most.
- The longest pass-to-table distance, about nine metres, adding carryover cooking that the spec never accounted for.
- The highest proportion of new starters, who followed the written spec exactly. Experienced grill chefs at other sites had quietly started pulling the chicken early because it looked done — they were compensating for a bad spec without ever reporting it, which is both impressive and the reason nobody found out for four months.
So Brierfield was functioning as the estate's most sensitive instrument. Weakest equipment, longest route, least improvisation — it registered a group-wide fault first and loudest.
The outlier was the detector, not the defect.
What the two plans would have cost
The original plan: two days on site, retraining the grill section, new probes, a GM performance conversation. Perhaps £3,000 in time and equipment.
It would have worked, for about five weeks. Retrained staff following a wrong spec revert to the spec. And the ten sites quietly drifting upwards would have carried on drifting, arriving at the same complaint level by roughly November, by which point the cause would have been four supplier changes and a menu revision in the past.
What happened instead: the spec was retimed in a test kitchen in two days — down to four minutes fifteen a side — reissued across the estate, and the grill boards reprinted.
| Aug | Oct | |
|---|---|---|
| Brierfield | 2.7 | 0.5 |
| Group | 0.6 | 0.1 |
Brierfield did not reach the group average. It settled at roughly five times it, which is about where its equipment and layout put it, and that is now a known and reasonable number rather than a crisis. The grill was replaced in the next capital round on the strength of a figure rather than an argument.
The check, written down
It is fifteen minutes and it belongs before every site visit:
- Take the theme the outlier is producing.
- Pull that theme's rate, per hundred covers, for every site, for the last four months.
- Look at the group row, not the site rows.
- Group flat, one site rising: local. Go and fix the site.
- Group rising, one site rising faster: shared. Find the input — a supplier, a spec, a menu change, a piece of software, a central policy — and check what changed around the month the rise started.
Step five has a shortcut that works embarrassingly often: ask procurement and marketing what changed, by date, in the last four months. The list is usually short and the answer is usually on it.
What the check does not do
Rates, not counts. On raw counts Brierfield was not the top of the list — the flagship was, because it does three times the covers. Ranking on counts sends you to your busiest site every quarter. The argument for the denominator in full, including why sites under about 400 covers a month are too noisy to rank.
Four months, not one. A single month's snapshot shows Brierfield at 2.7 and the group at 0.6 and says outlier. The direction of the group row is the entire finding, and one month has no direction.
Not every rise is shared. They ran the same check the following quarter on a rise in conduct complaints at Shawcross; the group row was flat and it genuinely was that site. The check is worth running because it is cheap, not because the answer is always "it's the group".
Compensation hides faults for a long time. Four months of experienced chefs silently pulling chicken early is four months of a spec being wrong without anyone being told. Worth asking directly at every site visit: what do you do differently from the spec, and why. The answers are a defect list nobody has written down.
You still have to decide where the Tuesday goes. This check tells you whether to drive anywhere. It does not rank three sites that all look wrong — that needs a scoring sheet with the weights written down in advance — and arriving without a specific thing to observe wastes the visit regardless. The version of a site visit that is not a coffee with the GM is the companion to this.
The thing that nearly cost them a quarter was not a lack of data. The rates existed, by site, by month, the whole time. What was missing was reading along the bottom row before booking the visit — which OMMU does by holding theme rates per site and flagging when a group row moves, and which a spreadsheet does equally well on the Monday somebody remembers to look.