Here is a league table of complaint rates per hundred covers from a fourteen-site group, worst at the top:
| Site | Complaints per 100 covers |
|---|---|
| Airport Retail | 1.9 |
| Westgate Drive-Thru | 1.6 |
| Station Concourse | 1.5 |
| Meadow Park | 0.7 |
| Church Lane | 0.6 |
| … | … |
| Old Mill | 0.3 |
Three sites are miles worse than everyone else. They have been at the top of this table every month for two years. Two area managers have run improvement plans on them. Neither plan moved the number.
They are not worse. They are different, and the table cannot tell the difference.
Why the same operation produces different rates in different formats
Four structural factors move a complaint rate substantially, and none of them is about how well a site is run.
Transaction speed and expectation. A guest in a station concourse has a train. A guest at a Sunday roast has an afternoon. The concourse guest's tolerance for a four-minute wait is near zero and the pub guest barely registers twenty minutes. Identical performance, different complaint rates.
Channel mix. Off-premise generates complaints at two to four times the rate of dine-in, for a mechanical reason: the guest cannot be recovered in the moment. Nothing can be comped, apologised for, or replaced before the bag leaves. A site at 60% delivery and a site at 5% delivery are not running the same business, and comparing their overall rate compares their channel mix.
Guest familiarity. Regulars complain less and forgive more. A site with 70% repeat custom and a site serving mostly first-timers — airports, stations, anywhere near a hotel — will differ by a wide margin on nothing but that.
Daypart. Breakfast and lunch produce fewer complaints per cover than dinner, partly because the ticket is simpler and partly because expectations scale with spend. A lunch-weighted site looks good on a blended number and may well be underperforming within its own daypart.
Put those together and a drive-thru at 1.6 may be outperforming a destination pub at 0.6. That is not a rhetorical flourish; it is why the improvement plans failed.
Building peer groups
You need enough groups to be meaningful and few enough to have more than one site in each. Most estates land on three or four, split on the factors above in roughly this priority:
1. Split by channel mix first. It is the biggest single driver. Anything above about 40% off-premise goes in its own group regardless of everything else.
2. Then by transit versus destination. Does the guest have a train, a flight, a school run — or an evening? This is the second-biggest factor and it tracks nothing else reliably.
3. Then by daypart weighting. Lunch-dominant separate from dinner-dominant.
4. Format and footprint last. Drive-thru, counter service, table service. Useful, but usually correlated with the first three, so it rarely creates a new split on its own.
For the fourteen-site group above:
| Peer group | Sites | Group mean |
|---|---|---|
| Transit and travel | Airport Retail, Station Concourse, Interchange | 1.6 |
| Off-premise weighted | Westgate Drive-Thru, Riverside, Hollins Rd | 1.4 |
| Suburban table service | Meadow Park, Church Lane, + 4 | 0.6 |
| Destination and events | Old Mill, Barn, Grange | 0.3 |
Read within those groups and the table says something entirely different. Airport Retail at 1.9 against a transit mean of 1.6 is modestly behind its peers — a real but small gap. Station Concourse at 1.5 is ahead of its group. And Church Lane at 0.6, sitting comfortably mid-table in the blended league, is exactly on its peer mean, which means two sites in that group are better than it and nobody had noticed because the number looked unremarkable.
The sites that were being managed hardest for two years were roughly at par. The ones quietly drifting were in the middle of the old table.
The rule that makes this work
Never compare a site to the estate. Compare it to its peer group, and compare the peer group to itself over time.
Two numbers per site, then:
- Gap to peer mean. Positive or negative, in complaints per hundred covers. This is the only fair measure of the site's management.
- Peer group trend. Where the whole group is going, month on month. This catches the thing a site-level view never does — a supplier change, a central menu decision, a policy that quietly broke something everywhere. The group row is where the real finding usually is.
Themes need the same treatment, and some need it more
Overall rate is the easy case. Theme-level comparison is where peer grouping earns its keep, because some themes are heavily format-dependent and some barely are.
| Theme | Comparable across the estate? |
|---|---|
| Cleanliness of toilets | Yes — compare everywhere |
| Staff conduct | Yes — the most portable theme you have |
| Order accuracy | Within channel only |
| Wait and pace | Within peer group only |
| Missing items | Off-premise peers only |
| Value and price | Within peer group and price band |
| Noise | Barely comparable at all |
Conduct and hygiene are close to universal. A rude greeting is a rude greeting in an airport or a barn, and if one site is producing conduct complaints at three times the estate rate that is real wherever it sits. These are your free comparisons and worth running estate-wide.
Pace and value are the ones people compare most and should compare least.
How peer grouping gets abused
Peer groups of one prove nothing. If you have a single airport site, it has no peers and the honest comparison is against its own trailing twelve months. That is a weaker instrument and pretending otherwise by putting it in a group it does not belong to is worse.
Peer groups become an excuse. The failure mode is predictable: every underperforming site discovers it is uniquely circumstanced. The defence is to set the groups once, on structural facts — channel share, daypart share, transit or destination — and then leave them alone for a year. If a site argues its way out of a group mid-quarter, you no longer have a measurement system.
You still need the denominator. Peer grouping fixes comparability; it does not fix volume. Raw counts will send you to your biggest site forever regardless of how well you group. Complaints per hundred covers first, peer groups second.
Small sites remain noisy. A site doing 300 covers a month moves a whole band on two complaints. Group them correctly and they will still swing; use a rolling three months for anything under about 400 covers.
Grouping does not decide your Tuesday. It tells you which gaps are real. Choosing between three sites with real gaps is a different exercise and benefits from weights you wrote down while calm — a scoring sheet does that.
The quarter-hour that pays for itself
Write your peer groups down. Today, on one page, with the split criteria named next to each group so that next quarter's argument has already been had.
Then rebuild your league table as four small tables instead of one big one. Most operators who do this find that at least one site they have been managing intensively is at par, and at least one they have never worried about is the genuine problem — which is a better return than almost anything else available for fifteen minutes.
OMMU holds peer groups and reports gap-to-peer-mean alongside the group trend, so the comparison is right by default rather than right when somebody remembers. A spreadsheet with a peer group column does the same job; what does not work is a single ranked list, however good the underlying numbers are.