The arithmetic of a rating: why volume is the only lever

The article accompanying this one deals with the listing: claiming it, what anybody can edit, the photographs, the duplicates, what can be reported, how to reply. It deals with the object.

This page deals with the number displayed on it, and with a property of that number almost nobody holds in mind: a mean behaves differently depending on how many values it aggregates, and that difference decides everything else.

The practical consequence is blunt. At twelve reviews, one one-star review costs several tenths of a point and shows in the search result. At two hundred, the same review moves the mean by a hundredth and nobody sees it. This is not a question of severity, it is a division.

So this article is about what that implies: why volume is the only real lever, what it actually takes to raise a rating, when to ask, what must never be done, and why a perfect score is a problem rather than a target.

Almost every satisfactory business has few reviews, and the reason is not dissatisfaction: it is that happy people do not think to write. Dissatisfaction motivates spontaneously.

That imbalance produces most unfairly low ratings. A business serving two hundred people a month well and asking none of them mostly receives reviews from those who had a problem — three a quarter, all negative, on a tiny base.

Asking corrects that bias, and it is the entire mechanism. It is not about obtaining good reviews but representative ones, which mechanically produces a higher rating if the service is sound — and a justly low one if it is not, which is information you want to have.

Regularity matters more than intensity. Five requests a week for a year produce a solid base and a flow of recent reviews; fifty requests in one week produce a visible anomaly and nothing afterwards.

So set a modest, sustainable number attached to an existing operation rather than to good intentions: at every delivery, at the end of every job, at every till transaction on a given day.

A reply to a review is not there to convince its author, who has rarely changed their mind, but to be read by the people who come afterwards — and they are far more numerous.

That changes what to write. A reply that rebuts point by point is addressed to the author; a reply that briefly explains what happened and what was done is addressed to the reader, and it is the one that works for you for two years.

Length is a signal in itself. A long reply to a short review gives the impression of a defensive business, whatever its content. Three lines is almost always enough, and the length constraint mechanically improves the tone.

Reply to positive reviews too, briefly and without a repeated formula. This is not politeness: a listing where each review has a specific reply shows a business that reads, which is exactly what somebody hesitating is looking for. Twelve instances of "Thank you for your trust!" produce the opposite effect.

One absolute rule: never reply the same day to a review that angers you. The delay costs nothing — a review is not an incident, it does not spread — and the reply written the next day is consistently better.

Many businesses aim for five out of five, and it is a counter-productive target worth understanding.

A perfect score over a significant volume reads as improbable, and it is: any activity serving hundreds of people produces dissatisfied ones. The reader infers either a small volume or filtering, and both reduce the trust they place in the whole.

Middling reviews in fact play a useful role nobody credits them with. A well-written three-star review, with a measured reply from the business, is often the most persuasive element on a listing — it proves the reviews are not filtered and it shows how you react when things go wrong.

The comfortable zone therefore sits lower than people think, and it is paired with a volume: a rating between 4.3 and 4.7 over several dozen reviews, with recent ones, is more solid than a 5.0 over twelve.

That frees you from two expensive behaviours: panic after an isolated negative review, and effort spent trying to remove three-star reviews, which are generally the most useful items on the listing.

What we do is bounded. We set up the review request at the right moment in your operation, we build the short link or code that removes the friction, we draft the replies, and we record the four quarterly figures with you.

We refuse to offer anything in exchange for a review, including a modest discount and including when a competitor is visibly doing it. It is against the platform’s rules, it is detectable, and the risk falls on the whole listing — that is, on years of legitimate reviews.

We also refuse to ask only the customers expected to give a good rating. It is the most widespread practice on this list and the most tempting; it removes precisely the information collection was meant to produce, and a business that filters its requests ends up not knowing what is wrong with it.

And we do not promise a rating. Regular collection produces a representative rating, which is good news when the service is good and bad news when it is not — in the second case what we recommend is fixing the problem first, because collection only accelerates the arrival of the truth. There is also something only you can do today: ask the next satisfied customer for a review yourself, at the moment they thank you.