Most complaints describe an outcome: the food was cold, the wait was long, the bill was wrong. Outcomes tell you something already happened.
A smaller number describe a condition — the state of the room, the demeanour of the staff, the way the service was being held together. Those are worth separating out, because a condition tends to precede a run of outcomes rather than summarise one.
Seven of them, with a caveat at the end that matters more than the list.
1. "The staff seemed stressed"
The single most useful phrase in this category. A guest who describes staff as stressed, rushed, frazzled or harassed is reporting that the floor was operating past its capacity — and they are reporting it from a visit where nothing necessarily went wrong for them.
Under-capacity shifts produce a small number of failures at random. One guest noticing the strain often precedes several noticing the consequences.
2. "They were clearly short-staffed"
Related and more explicit. Note that this is a guest's inference, which makes it more interesting than a complaint — they have looked at the operation and diagnosed it. Guests are usually right about this and it is worth checking against the rota rather than dismissing.
3. "One poor girl was running the whole section"
A named or implied individual carrying a service. Almost always accurate, and predictive in a second way: it identifies a retention risk. The person being described is the one you are most likely to lose, and losing them makes the condition worse.
4. "Nobody seemed to know what was going on"
A coordination failure rather than a capacity one. Points at a handoff, a missing shift leader, or a new manager without the floor's confidence. Distinct from being short-staffed and needing a different response — this is where the shift handoff audit is the right tool.
5. "It was chaos when we arrived"
Door and pacing. Frequently a booking configuration problem rather than a staffing one: too many covers seated in the same fifteen minutes, which guarantees a bad first twenty and a compressed kitchen. Worth checking against the booking intervals before assuming a people problem.
6. "Not what it used to be"
The most serious on the list and the slowest. It is a comparison from someone with an accurate memory, and it never refers to a single visit. Regulars are the last group to complain and the first to leave, so this phrase tends to appear when a decline is already established.
Three of these in a quarter is worth taking more seriously than thirty complaints about waiting.
7. "We won't be rushing back" — with nothing else wrong
A retention signal with no attached fault. There is nothing to fix, which is what makes it easy to dismiss, and it is the phrase most correlated with the quiet decline you notice in the covers six months later. Related to the almost-happy guests discussed in the four-star retention piece.
The caveat, which is the important part
These are not predictions. They are conditions that make failures more likely, and the honest position is that the evidence for treating any single one as an early warning is weak — review volumes are far too low for one comment to carry predictive weight. One "staff seemed stressed" is one guest's impression on one evening.
What they are genuinely useful for is changing what you look at. A cluster of condition phrases in a fortnight is a reason to check the rota and the booking intervals for that daypart, which is a cheap thing to do and occasionally finds something real. It is not a reason to act on the phrase itself.
The failure mode is exactly the one in the Friday pattern that was not real: an operator finds a story in a handful of comments, spends money on it, and the comments continue because they were never the signal they appeared to be. Three mentions of stressed staff is a prompt to look at a rota. It is not a finding.
How to use the list properly
Keep them as a tagged category, separate from complaints, and watch the count rather than the individual instances. Two things then become visible that are otherwise invisible:
A rise in condition phrases without a rise in outcome complaints. This is the genuinely useful pattern — the service is straining and has not yet failed visibly. It is the only version of this that deserves the word leading.
Condition phrases concentrated in one daypart. Much stronger than a general rise, and it points at a specific shift with a specific rota.
Both need a baseline, and both need the phrases to be tagged consistently over months rather than noticed in passing. Nobody maintains that by hand — the comments are too spread out and too easy to read as one-offs. It is what OMMU counts in the background, so a fortnight where the condition language doubles arrives as a note rather than as a feeling that something is off.