A star rating is not moved by working on reviews. It is the delayed output of how a property is run. We operate three buildings and six rooms in Tokyo and Sapporo, all on hotel-business permits (旅館業許可, the Japanese licence with no annual cap on nights), and across all listings we hold 364 reviews at a weighted 4.9 (measured 21 August 2026, operating data).
One team, one set of manuals, one cleaning standard, and the per-room averages still range from 4.81 to 5.00 (as of 31 August 2026, published live on each stay page). Where that 0.19 comes from is the subject of this article.
How is the rating actually set?
It is the mean of every past review, which makes it asymmetric: one night lowers it, months raise it back.
When one low review lands on a listing sitting at 4.9, the number of later five-star reviews needed to bring it back to 4.9 does not depend on how many reviews that listing already holds. It is worth checking on paper, because it decides the whole strategy.
| The review that lands | Fives needed to return to 4.9 | In practice |
|---|---|---|
| One 4 | 9 | more than a month of one room's arrivals |
| One 3 | 19 | close to three months |
| One 2 | 29 | approaching half a year |
| One 1 | 39 | most of a year |
Why is a new listing so fragile?
Because the movement of any single review is divided by the reviews already there. The same 3 moves a room holding 126 reviews by about 0.015, and a room holding 33 by about 0.06.
Our arrivals run at an average stay of 3.3 nights and 449 bookings over twelve months, about seven per room per month (measured August 2026, operating data). Even if every guest wrote, one 3 consumes close to three months of that room's output. So a rating is not a thing you raise, it is a thing you stop lowering, and removing one reason it falls beats any campaign to enlarge the denominator.
Why do rooms run by the same team differ?
Less because of variation in the cleaning than because of what the building brings with it and how old the listing is (as of 31 August 2026, published on each stay page).
| Room | Reviews | Average |
|---|---|---|
| SAPPORO NORTH 1A | 33 | 5.00 |
| KAMEARI 101 | 42 | 4.98 |
| SAPPORO NORTH 1B | 53 | 4.91 |
| KAMEARI 201 | 54 | 4.81 |
| SAPPORO NORTH 1C | 70 | 4.83 |
| HEIWAJIMA | 126 | 4.87 |
The three rooms with the most reviews sit at the bottom and the three youngest sit at the top. An average carries its opening months forever: those faults have been fixed everywhere except in the mean. A young listing, meanwhile, has simply not met its first bad night yet.
The building contributes too: whether the entrance is shared, whether there are stairs, what the street sounds like at night, and how closely the room matches the impression the photographs give. Our lowest-rated room is not the one cleaned worst. It is the one where the gap between expectation and fabric opens widest.
What moves the number most?
In order: cleanliness, accuracy of the listing, speed of reply. Very little outside those three reaches the stars at all.
Cleanliness comes first because a guest judges it within ten minutes of walking in and the judgement is binary. Cleaning is our largest cost line at 17.5% of revenue precisely because it is the floor the rating stands on, which is why it is the wrong line to cut (how cleaning runs). Reply speed matters on the bad days rather than the ordinary ones: a night spent stuck in front of a door lands in a review with a different weight from a missing kitchen item (24-hour guest response).
How honest should the listing text and photos be?
A rating is the gap between expectation and reality, so the cheapest improvement available is not raising the reality, it is lowering the promise.
Our listings state what is absent as well as what is there. The three Sapporo kitchens have a rice cooker, a stove, a microwave and a fridge, and no oven. At Kameari the ground-floor room 101 is stairs-free and room 201 is up a flight. Left unsaid, both become discoveries made on arrival. Photographs go out in the order the rooms are actually met. Writing the listing bolder to win one more booking, and losing 0.02 off the average afterwards, are the same operation seen from two sides.
Does asking guests for five stars work?
We do not ask. Asking enlarges the denominator, not the reason the number fell. If a second 3 arrives for the same reason before those nineteen fives accumulate, you are back at the start, and removing the cause makes the nineteen unnecessary.
Send a request around checkout and what we collect is a rating while what the guest collects is homework. We keep a channel open around the clock to reach somebody who is struggling, not somebody who is leaving. Each booking platform also publishes its own policy on reviews and incentives in its help pages, and revises it, so check the current version there before designing any process around reviews. Anything stated here as a general rule would start ageing the day it was written.
What we do instead is publish what arrives. The reviews on a stay page are the most recent ones, not the best ones. As of 31 August 2026, two of the eight most recent reviews on one of the Sapporo rooms are 4s, and they are on the page (SAPPORO NORTH stay page).
A low review lands. What happens?
The physical thing gets fixed before the reply gets written. Do it the other way round and the same review arrives again.
Our order is four steps. Identify which layer it belongs to: cleanliness, listing accuracy, getting in, or response. Fix whatever is physically fixable at that week's turnover. Reply in public, short and factual. And if the cause is a condition of the building that cannot be fixed, rewrite the listing so the next guest is not surprised by it.
Replies stay short because the audience is the person reading while deciding whether to book, not the person who wrote it. The fourth step does the most work: keep an unfixable condition out of the listing and reviews for that same reason keep arriving on a schedule, taking the average apart from underneath.
How should an owner read a management company's ratings?
Ask to see the unselected version rather than one good number, because the average of one strong room may be telling you about that building rather than that company. Four questions are worth putting: the average across every room they run, whether they will show the last ten reviews unselected, how many minutes a night message takes to reach a person, and whose account holds the listing and its review history (switching management companies).
And here is what we will not promise. We do not tell an owner what their property will score. A rating is a composite of the building, the location and the operation, and an operator owns only part of it. What we commit to is our part: the cleaning standard, the response at night, and the accuracy of what the listing says.