An area manager reports 94% response coverage. The site in question has answered 47 of 112 reviews in the period.
Both figures are honest. The area manager counted replies to reviews of three stars or below, in the last 30 days, excluding reviews with no text. The 42% figure counts every review on every platform for the quarter. Neither definition is wrong and the two numbers cannot be used in the same sentence.
This is the normal condition of review metrics in hospitality. The numbers get quoted, compared across sites, put in board packs and used to judge people, and the definitions live in somebody's head. So here are five, written as formulas, with the decisions that have to be made explicit.
1. Review velocity
Not total review count, which is a measure of how long you have existed.
Velocity = reviews received in the last 30 days
And the version that is actually useful for comparison:
Velocity index = (reviews in last 30 days) ÷ (mean monthly reviews over the preceding 12 months)
An index of 1.0 is normal for you. 0.6 means your review flow has dropped by 40%, which is worth knowing because it usually means the asking stopped rather than that anything went wrong.
Decisions to write down: 30 days or calendar month (use 30 days — calendar months have different lengths and that alone moves the number by 10%). Which platforms. Whether a review edited this month counts as this month's.
Normalise it for volume if you compare sites:
Reviews per 1,000 covers = (reviews ÷ covers) × 1,000
This is the only version that lets a 200-cover site and a 2,000-cover site be compared, and most estates have never calculated it.
2. Response coverage
The metric most often quoted and least often defined.
Coverage = replies posted ÷ reviews eligible for reply
Everything depends on eligible, and you must choose and then never change it:
| Decision | Options | Recommendation |
|---|---|---|
| Star range | All, or ≤3 only | All. A four-star review is the most valuable one to answer |
| Text requirement | All, or text-only | Text-only. A bare star rating cannot be meaningfully replied to |
| Window | Rolling 30 days, or quarter | Rolling 30 days, lagged by 7 |
| Platforms | All, or Google only | All you have access to, listed by name |
The seven-day lag is the detail that matters and almost nobody includes it. If you measure coverage up to today, the reviews from the last few days have not had a fair chance of being answered, and the metric punishes you for recency. Measure coverage for the period ending a week ago.
Keep it separate from speed, which is its own number:
Median hours to first reply
Median, not mean — one review answered three weeks late will drag a mean into uselessness while the median tells you what normally happens.
3. Theme rate
The most operationally useful of the five, and the one with the most ways to go wrong.
Theme rate = (reviews mentioning the theme ÷ total reviews in period) × 100
That gives you a share of reviews, which is right for reading how you are described. For anything operational you want the other denominator:
Theme rate per 100 covers = (reviews mentioning the theme ÷ covers) × 100
These answer different questions and both are legitimate. Share-of-reviews tells you what a stranger reading your profile will come away thinking. Per-hundred-covers tells you how often the thing is actually happening, which is the only version that can be compared across sites or across months when volume changes. The argument for the covers denominator is worth reading in full.
Decisions to write down: your theme list, fixed, in writing — eight to twelve themes, and do not add one mid-quarter or your series breaks. Whether a review can count in two themes (yes, it can and should). Whether positive mentions count (keep them separate; "friendly staff" and "rude staff" are not one theme).
4. Profile accuracy
The only one of the five that is not derived from reviews, and the only one that is binary.
Accuracy = fields correct ÷ fields checked
Define the field list once — regular hours, special hours for the next 90 days, phone, address, menu link resolving on mobile, service-option attributes, payment attributes, most recent photo under 90 days old — and check it monthly. Eight fields, score out of eight.
It is a crude metric and it earns its place because it is leading rather than lagging. Every other number here tells you about something that already happened. A wrong special-hours entry tells you about reviews you have not received yet.
5. Outlier gap
For anyone running more than one site.
Gap = site theme rate − peer group mean theme rate
Expressed in the same units as the theme rate, signed, so that negative is better than peers.
The word doing the work is peer. A gap against the estate mean is not interpretable, because a transit site and a destination pub do not share a baseline. Peer groups have to be set before this metric means anything.
And always alongside its companion:
Peer group trend = peer group mean this month − peer group mean three months ago
A site with a big positive gap and a flat peer trend is a local problem. A site with a big gap and a rising peer trend is usually the most sensitive detector of something group-wide — which is a completely different quarter.
The two that look like metrics and are not
Average star rating. Not useless, but nearly unmanageable: it is cumulative, so with 600 historical reviews your current month barely registers, and the number you can affect is buried under years of arithmetic. If you must track it, track the mean rating of the last 30 days' reviews as a separate figure, which moves and is therefore readable.
Sentiment score. A single number compressing all your review text into a value between −1 and 1. It feels precise and nothing actionable can be recovered from it. A sentiment score that falls from 0.42 to 0.31 does not tell you to do anything; a noise theme rate rising from 11% to 24% does.
Writing the definitions down
One page, in the shared drive, named how we calculate review metrics, with a version date.
For each metric: the formula, the denominator, the window, the platforms, the lag, and the date you last changed it. Then leave it alone for a year.
That last part is the discipline. Changing a definition mid-series destroys the series, and a series is where all the value is — one month's theme rate tells you almost nothing, and twelve months of a consistently-calculated one tells you whether anything you did worked. An inferior definition held constant beats a better one that changed in June.
The failure modes of all five
Small numbers break all of them. At 15 reviews a month, a theme rate moves 7 percentage points on one review. Below about 25 reviews a month, use rolling 90-day windows throughout and resist the urge to read monthly movement.
Platform coverage is uneven. If one site is on three platforms and another on one, their velocity and coverage are not comparable. Either restrict to the platform everyone has, or normalise — and say which you did.
Metrics attract gaming once they are used to judge people. Response coverage is the most gameable here: a one-word reply counts identically to a good one. Watch review volume at any site whose rate improves suddenly — flat or falling volume alongside an improving rate is the signature of a site managing the number.
None of these is a target. They are instruments. What good looks like is a question about your own history rather than an industry figure, and published benchmarks mislead more than they help.
OMMU calculates all five on fixed definitions with the denominators visible, which removes most of the arguments above. A spreadsheet does it equally well — the thing that is not optional is the one page saying what each number means, because the alternative is two honest people reporting 94% and 42% for the same month.