The usual argument for paying attention to four-star reviews is that they will lift your average. That argument is weak, and it is worth abandoning before someone in your business checks it.
Two hundred reviews at 4.2 need roughly sixty consecutive five-stars to reach 4.4 — the same arithmetic that makes reply rate the better target than star rating. Converting four-star reviewers to five-star reviewers is a slow and largely invisible way to move a number that guests barely calculate.
The real case is retention, and it is much larger.
The calculation
Suppose 500 covers a week at £34 average spend. Suppose — and this is the assumption doing the work — that one guest in twelve leaves in the state a four-star review describes: satisfied, mildly caveated, undecided about returning.
That is around 40 guests a week sitting on the fence, most of whom never write anything at all. The reviews are just the written minority of a much bigger group.
Recover five of them per week into one additional visit a year each:
- 5 × 52 = 260 extra visits a year
- At £34, that is roughly £8,800 of revenue
- At a 70 per cent gross margin, about £6,200 of contribution
For fixing an acoustic problem or trimming a menu section. And unlike a discount, none of it comes out of the price.
The numbers are illustrative — your average spend, party size and margin will differ, and the one-in-twelve is a judgement rather than a measurement. Redo it with your own figures before quoting it at anyone. The shape holds regardless: the value is in the returning visit, not in the star.
Why the ratings barely move even when this works
Worth being clear about, because it is the thing that makes people abandon the project after two months.
A guest recovered into returning does not usually write a second review. Review writing is mostly a first-visit behaviour, triggered by novelty or by something going wrong. So a successful retention improvement produces almost no visible change in your review profile for a long time.
What changes first is covers on your quieter shifts, and repeat visit rate if you have any way to see it. Both are till-side measurements. If you set up this work with a ratings target you will conclude it failed while it is working, which is the standard way these projects die.
Which caveat to work on
Not the most common one. The most reversible one — the caveat where a fix plausibly changes the return decision.
| Caveat | Frequency | Does fixing it bring them back? |
|---|---|---|
| Room too loud | Often | Yes — it is why they chose elsewhere |
| One dish underwhelming | Often | Rarely — they will order something else |
| Bit pricey, nothing else wrong | Often | No, not without a positioning change |
| Slow between courses | Medium | Yes |
| Hard to decide what to order | Medium | Indirectly, via spend rather than return |
| Table cramped, seat uncomfortable | Low | Yes, strongly, and cheap to fix |
The bottom row is the pattern worth noticing. Physical comfort complaints are rare in reviews and disproportionately decisive, because a guest who was uncomfortable for ninety minutes remembers the discomfort and not the reason. Low frequency, high reversibility. Meanwhile "bit pricey" is frequent and mostly not reversible by anything except a genuine positioning decision.
Sorting caveats by frequency alone points you at the wrong work.
What to actually do
Get the caveat counts first — the extraction method is in the four-star piece. Then score each theme for reversibility rather than volume, pick one, fix it properly, and measure on covers rather than stars.
Give it a quarter. Retention effects are slow by construction: a guest whose return interval is eleven weeks cannot demonstrate anything in a fortnight, and a single week proves almost nothing regardless of what it appears to show.
The awkward requirement is that the caveat counts need to still exist in three months so you can compare. That continuity is the whole practical difficulty, and it is what OMMU maintains — the themes stay counted quarter over quarter, so "noise mentions are down by two thirds since March" is a sentence someone can actually say.