Raising an average rating is arithmetic on a mean, not a campaign. Everything already rated sits in the numerator and in the divisor at once, so each new rating dilutes the old ones only a little. New ratings must also be strictly better than the target you picked, or that target stays out of reach at any volume whatsoever.
An average rating is an arithmetic mean, and a mean has inertia
The number under your venue name on a maps listing is not a mood, a score or a reputation index. It is the arithmetic mean of every published rating, recalculated after each new one. That is the whole mechanism, and it is worth stating plainly, because almost every wrong expectation about ratings comes from treating the number as something more elaborate than division.
Average rating — the arithmetic mean of all published ratings of one venue on one platform.
Once you accept the division, the consequences follow without any theory. A mean carries its own history: the ratings you collected two years ago are still in the sum, and they are still counted in the divisor. Nothing you do today removes them. You can only add to both sides of the fraction, and adding to a large divisor is slow work.
Rating base — the number of ratings already counted in the average. It is what makes an average heavy: the same new review moves a base of fifty far more than a base of five thousand.
Rating inertia — the effect of a single new rating on the average, which falls as one over the base plus one.
The base is the variable owners forget when they set a goal. Two venues with the same 4.3 are in completely different positions if one of them has 40 ratings and the other 4,000. The first can change its number with ordinary effort. The second cannot, and no amount of enthusiasm will change that, because the obstacle is a divisor and not a lack of will. Before anything else on this page, write down two numbers from your own listing: the average you have now, and how many ratings it is built on. Both are visible to anyone, including you; if you want them tracked over time next to your other numbers, that is what a reporting layer is for.
What the average becomes after k new ratings
The first formula is simply the definition of a mean, written so that the old ratings and the new ones stay visible separately.
New average = (A × N + Σ rᵢ) ÷ (N + k)
A— current average rating on the platform, stars;N— current number of ratings, pieces;rᵢ— each of theknew ratings, stars;Σ rᵢ— the sum of the new ratings, stars;k— how many new ratings arrived, pieces.
The dimensions work out because A × N is not a rate times a count in any physical sense: it reconstructs the accumulated sum of stars that the platform has been carrying all along. Stars plus stars, divided by pieces, gives stars again.
Take the venue that will serve as the worked example on this page: 180 published ratings, average 4.3. The accumulated star sum is 4.3 × 180 = 774. That single number, 774, is your real starting position — not the 4.3, which is only what 774 looks like after division.
Now add ten five-star ratings. The sum becomes 774 + 50 = 824, the count becomes 190, and the new average is 824 ÷ 190 = 4.3368, which the platform will show as 4.3. Ten perfect ratings, and the visible number did not change at all. That is not a failure of whatever produced them. It is what division by 190 does.
How many ratings it takes to reach a target, and the hard condition attached
Reverse the first formula and you get the number most owners actually want. Assume for simplicity that every new rating is worth the same r stars — an assumption we will relax in a moment.
k = N × (T − A) ÷ (r − T), defined only when r > T
N— current number of ratings, pieces;A— current average, stars;T— target average you want to reach, stars;r— the rating each new review is worth, stars;k— how many such ratings you need, pieces.
Stars divide by stars inside the brackets, leaving a pure number that multiplies pieces: the result is pieces, which is what a count of reviews should be.
The worked example, all the way through
The venue has N = 180, A = 4.3, and wants T = 4.5. Suppose every new rating is a five, so r = 5.
k = 180 × (4.5 − 4.3) ÷ (5 − 4.5) = 180 × 0.2 ÷ 0.5 = 72
Seventy-two consecutive five-star ratings, with not a single four among them. Check it forward: (774 + 5 × 72) ÷ (180 + 72) = 1,134 ÷ 252 = 4.50 exactly. Check the step below it: 71 fives give 1,129 ÷ 251 = 4.498, which rounds to 4.5 on a display but is not 4.5. The formula does not overshoot; it lands on the target.
Now make the assumption honest. Nobody collects seventy-two consecutive fives. Suppose new ratings average r = 4.6 instead — still well above the current 4.3, still a good stretch of service:
k = 180 × 0.2 ÷ 0.1 = 360
Three hundred and sixty. Check: (774 + 4.6 × 360) ÷ 540 = 2,430 ÷ 540 = 4.50. Dropping the quality of incoming ratings by four tenths of a star multiplied the work by five. This is the first place where the arithmetic stops being intuitive, and it is worth sitting with before reading on.
The ratio that does not depend on your base
Divide both sides by N and something useful appears:
k ÷ N = (T − A) ÷ (r − T)
For our venue that is 0.2 ÷ 0.5 = 0.4. You need new ratings worth forty per cent of your entire rating history, whatever that history happens to be. A venue with 1,800 ratings at 4.3 needs 720 fives to reach 4.5 — the same forty per cent. The base does not change the proportion; it changes the absolute amount of work, which is the number you have to live with.
The condition r > T in plain words: ratings at or below your target never get you there
The formula carries a condition, and the condition is not a footnote. It is the single most useful sentence on this page.
If r equals T, the divisor is zero and there is no answer. That is not a technical inconvenience — it is correct. New ratings worth exactly your target pull the average towards the target and never past it; you approach 4.5 and never arrive.
If r is below T, the formula returns a negative number. For our venue with r = 4.4: k = 180 × 0.2 ÷ (−0.1) = −360. There is no such thing as minus three hundred and sixty reviews. The negative sign is the arithmetic telling you the target is unreachable with ratings of that quality, and it will stay unreachable no matter how many arrive. Every 4.4 you collect moves the average up a little and moves the ceiling of what you can reach down towards 4.4.
Say it once in the room and it saves an argument later: if the average rating of your new reviews is not strictly above your target, your target is not a target — it is a wish. Fixing that requires changing the service that produces ratings, not the volume of requests for them. This page holds the arithmetic; the working half of the subject — who answers, in what tone, within what limits — is a different job, described in the article on responding to Google reviews, and this page deliberately gives no advice about it.
The weight of a single rating: why a venue with a thousand reviews barely moves
There is a smaller formula hiding inside the first one, and it explains more day-to-day frustration than any other line here.
Effect of one review = (r − A) ÷ (N + 1)
r— the rating of the new review, stars;A— average before it arrived, stars;N— number of ratings before it arrived, pieces;- result — the shift of the average, in stars per one review. Print it to at least four decimals, or it will look like zero.
For our venue, one new five-star rating moves the average by (5 − 4.3) ÷ 181 = 0.0039 stars. One new one-star rating moves it by (1 − 4.3) ÷ 181 = −0.0182 stars.
The same review at four different bases
Keep A = 4.3 and r = 5, and change only the base:
- 50 ratings — one five moves the average by 0.0137 stars;
- 180 ratings — 0.0039 stars;
- 1,000 ratings — 0.0007 stars;
- 5,000 ratings — 0.0001 stars.
The strongest reading of that list is not that big venues are stuck. It is that the value of a review depends on when you collect it. Ratings gathered while your base is small are worth an order of magnitude more than the same ratings gathered later, and no later effort recreates that. A venue that starts asking guests for ratings early is not being diligent; it is buying leverage that stops being available. The asking itself is a process — a message after the visit, sent to the right guest and not to everyone — which is where an automated post-visit message and a guest record do the work that arithmetic cannot.
One one-star among fives: how many fives put the mean back
Notice the two weights above: 0.0039 upward for a five, 0.0182 downward for a one. The bad rating is worth 4.71 times the good one, and that number is not a coincidence. Divide the two weights and the base cancels out entirely:
Fives needed to offset one one-star = (A − 1) ÷ (5 − A), rounded up
A— the average you want to hold, stars;- the numerator is how far a one-star sits below that average, stars;
- the divisor is how far a five-star sits above it, stars;
- result — a plain count of reviews, no unit: stars divide by stars.
This is a special case of the previous formula, taken at the point where the average returns to where it was. Because the base cancels, the answer is the same for a venue with fifty ratings and for a venue with five thousand.
The ladder, and why it is the point of this page
Run the formula at several averages and round up:
- average 4.0 — one one-star costs you 3 fives;
- average 4.3 — 5 fives;
- average 4.5 — 7 fives;
- average 4.8 — 19 fives;
- average 4.9 — 39 fives.
That is the whole argument. Between 4.5 and 4.8 the visible number rises by three tenths of a star — about seven per cent — and the price of a single bad evening nearly triples. The higher your rating, the less room a five-star has to lift you and the further a one-star has to fall. Reputation does not get easier to defend as it improves; it gets arithmetically harder, and the curve steepens exactly where owners assume they have finally arrived.
Verify it by brute force rather than trusting the algebra: at an average of 4.5, add one one-star and then add fives one at a time until the average is back at 4.5 or above. The answer is seven at a base of 50, seven at 100, seven at 1,000 and seven at 5,000. At 4.8 it is nineteen at every one of those bases.
When the target is out of reach, and what to aim at instead
Put the two results side by side for our venue and the picture is complete. To go from 4.3 to 4.5 it needs 72 fives — forty per cent of its entire history. To go on from 4.5 to 4.8 it would need 180 × 0.3 ÷ 0.2 = 270 fives, or 150 per cent of the base it had at the start. Check: (4.5 × 180 + 5 × 270) ÷ 450 = 2,160 ÷ 450 = 4.80.
The comparison that decides it
Take your k and hold it against two numbers you already know: how many ratings you have collected in your whole history, and how many guests you actually serve. If k is larger than your entire rating history, you are asking for more new ratings than everything the venue has ever accumulated. If k is larger than the number of guests you serve in the period you are looking at, the target is not ambitious — it is arithmetically unavailable, and calling it a goal will only teach your team that goals are decorative.
What to aim at instead is a decision the formula can inform but not make. Three targets stay honest when the headline average will not move:
- the quality of new ratings,
r, which is the only input that changes what is reachable at all; - the share of guests who leave any rating, which turns
kfrom a number on paper into a count of requests; - the distribution behind the average — twenty fives and four ones is not the same venue as twenty-four fours, even though the mean is close.
None of those three is the number on the listing, and that is the point. Where the average is a lagging summary, they are the inputs. If you are assembling a set of numbers to watch rather than a single one, the KPI tree for a restaurant puts reputation in its place among the rest, and the numbers an owner actually looks at is the shorter version of the same argument.
Recency weighting: why this arithmetic is a lower bound on movement
Everything above assumes a plain mean in which every rating counts once, forever. Some platforms state that recent ratings weigh more than old ones; the weights themselves are not published, and this page does not guess at them.
Recency weighting — a platform rule that gives newer ratings more weight than old ones. Where it applies, the plain-mean arithmetic on this page is a lower bound on how fast the visible number moves.
The direction of the error is what matters, and it is favourable. If old ratings decay in weight, your effective base is smaller than the count shown on the listing, and a smaller base moves faster. So the k you compute here is the pessimistic version: the real number needed is that or less, never more. Building your plan on the plain mean means you will be wrong on the safe side.
Two practical consequences follow. First, do not try to reverse-engineer the weights from your own listing: with a base of a few hundred you cannot separate a weighting rule from ordinary variation in what guests type. Second, if your average moves noticeably faster than this page predicts, that is evidence of weighting rather than of a miscalculation — check the arithmetic once, then stop checking it. Where your rating is displayed, and which of your data the platform shows next to it, is a separate matter covered in restaurant data in Google, on your site and on portals; keeping that record correct is what a profile layer is for.
What this arithmetic does not solve: it says nothing about where ratings come from
The formulas above take k and r as given. They are not given. Everything difficult about ratings lives in producing them, and none of it is arithmetic.
Here is the boundary, drawn explicitly, because a formula presented without one invites the reader to believe it does more than it does.
| Action | Moves the average | Moves something else | Why |
|---|---|---|---|
| A new rating is published | Yes | — | It enters the sum and the divisor at the same time |
| You reply to a review | No | Whether the guest revises it, and what other readers see | A reply carries no stars |
| A guest edits their own rating | Yes | — | The old star value is replaced, not added to |
| The platform removes a fake rating | Yes | — | It leaves both the sum and the divisor |
| You serve more guests without asking for ratings | No | Revenue, occupancy, coverage | Guests who do not rate never enter the mean |
| You fix what people complain about | Not directly | The distribution of future ratings | It changes r, not A |
| A review is left as text with no stars | No | — | No stars, nothing in the sum |
Two of those rows deserve emphasis. Replying does not move the average — the reply is not a rating and never becomes one. And fixing the underlying problem does not move the average either, not on the day you fix it: it changes the quality of ratings that have not been written yet, which is r, and r is the input that decides whether your target exists at all.
isoc_cismpn2, indicator E_SM_PCUQOR, NACE section I, enterprises with 10+ persons employed, data updated 11 December 2025, opened 27 August 2026); the EU-27 figure for the same year is 53.18 %.Read the label carefully: it covers section I, accommodation and food service together, hotels included, and only enterprises with at least ten people employed. It is not a statement about restaurants, and it is not about star ratings — it is about the practice of collecting and answering opinions at all.
There is no official statistic to compare your average rating against. Neither the European statistical catalogue nor the Polish statistical office publishes venue ratings, review counts or reputation benchmarks of any kind, and this page therefore names no norm. Your comparison base is your own previous number and your own distribution — nothing else exists, and anything presented as an industry standard for average rating has no publisher behind it.
Whether a change in your rating is a real change or ordinary noise has its own method, and it is the subject of testing whether a promotion worked. Which guests are behind those ratings, and how many of them return at all, belongs to guest retention and repeat rate. Where your rating is read by machines rather than people is covered in the facts assistants read about you. And the operational half — requests, replies, complaint handling — is where review workflow and a single queue instead of five belong; the wider picture of what gets automated in a venue is in restaurant automation.
Frequently asked questions about rating math
How many reviews do I need to raise my rating from 4.3 to 4.5?
Multiply your current number of ratings by the gap you want to close and divide by how far your new ratings sit above the target. With 180 ratings and only five-star reviews arriving, that is 180 × 0.2 ÷ 0.5 = 72 reviews. With new reviews averaging 4.6 instead of 5, the same move takes 360. Substitute your own base — the count you already have matters more than the average you start from.
Why does my rating barely move when I get new reviews?
Because one review shifts the average by (r − A) ÷ (N + 1), and N sits in the divisor. With 180 ratings at 4.3, a new five-star rating moves the number by 0.0039 stars — a change no display will show. At 1,000 ratings the same review is worth 0.0007. Nothing is broken; the divisor is simply large.
How many five-star reviews cancel out one one-star review?
(A − 1) ÷ (5 − A), rounded up, and the answer does not depend on how many ratings you have. At an average of 4.0 it is three fives. At 4.5 it is seven. At 4.8 it is nineteen, and at 4.9 it is thirty-nine. The better your rating, the more expensive one bad evening becomes — that steepening is the real subject of this page.
Can a target rating be impossible to reach?
Yes, and the formula says so out loud. It only has an answer when the average of your new ratings is strictly above the target. If new reviews average exactly the target you approach it forever without arriving; if they average below it, the formula returns a negative count, which means the target is unreachable rather than that you need negative reviews.
Does the platform weight recent reviews more heavily?
Some platforms state that they do, without publishing the weights, so this page assumes a plain mean instead. That assumption errs in your favour: if old ratings lose weight, your effective base is smaller than the visible count and the number moves faster than calculated here. Treat every k on this page as the pessimistic figure.
Does replying to a review change the average?
No. A reply carries no stars, so it changes neither the sum nor the divisor. It can matter for a different reason — the guest may revise their own rating, and a revision does replace the old star value — but that is a separate mechanism with its own practice, and this page does not cover it.
Do the calculation once, with your own two numbers, before you set any target for the room. If k comes out larger than the number of guests you can realistically serve in the period you are looking at, the honest conclusion is that the target is wrong, not that the team is. Start from the restaurant section if you want the rest of the arithmetic that sits around this one.