At our own three buildings and six rooms in Tokyo and Sapporo, 711 confirmed bookings were due to arrive between November 2025 and October 2026, and 196 of them (27.6%) were later cancelled (counted on 16 November 2026 from the records in Hospitable, the reservation system we run the rooms on).
A cancellation policy is chosen before the first booking ever arrives, which is exactly when you have no data. This article works backwards from when cancellations actually came to which refund tier matters in practice.
What is a cancellation policy, and what does it actually decide?
It is the rule for how much a guest gets back if they cancel, depending on how many days before arrival they do it. Each booking platform offers a set of tiers, and the host picks one per listing.
It decides more than the refund. How far ahead a date reopens decides whether it can be sold again. Tier names and refund rates differ between Airbnb, Booking.com and the rest, and they change, so this article deliberately describes none of them: check each platform's help pages for the current terms before you choose.
How often are bookings really cancelled?
For us, roughly one confirmed booking in four: 27.6%. The two main platforms are close to each other.
| Channel | Confirmed bookings | Later cancelled | Share cancelled |
|---|---|---|---|
| Airbnb | 625 | 170 | 27.2% |
| Booking.com | 85 | 26 | 30.6% |
| All channels (incl. direct) | 711 | 196 | 27.6% |
How far before arrival do cancellations come?
Mostly early. The median of the 196 was 37 days before arrival, and 58.2% came 30 or more days out.
| When the booking was cancelled | Bookings | Share |
|---|---|---|
| 30 days or more before arrival | 114 | 58.2% |
| 7 to 30 days before | 62 | 31.6% |
| 1 to 7 days before | 14 | 7.1% |
| Less than a day before | 6 | 3.1% |

Which refund tier actually makes a difference?
The one covering the last 30 days, which is where four in ten cancellations fell. Our bookings arrive a median 40 days before the stay (operating data), so a date that reopens 30 or more days out goes back on sale before close to half of its demand has even turned up.
A date that reopens inside the last week has almost no time left to sell. That is where a stricter refund earns its keep: the 10.2% inside seven days, and to a lesser degree the 31.6% before that. Whether a stricter policy also puts guests off booking in the first place cannot be measured without running both side by side, so we have no number for it and do not pretend to.
One more figure: 24.5% of cancellations came within 24 hours of the booking being confirmed. We read most of these as someone still deciding, rather than a lost stay.
Which terms do we use ourselves?
On our own direct booking channel: a full refund within 24 hours of booking or up to 30 days before arrival, 50% up to 7 days before, and nothing after that (as set on 16 October 2026, identical on every listing).
The logic is the timing table above. A cancellation that leaves time to resell costs the guest little; one that leaves no time costs more. On each platform we pick whichever of its own tiers comes closest to that shape, and re-check the settings and the help pages when a platform changes its terms.
What does the operator do on the day a booking is cancelled?
First, confirm the dates have reopened on every channel, because they do not reopen at once. In our own test on 16 October 2026, after a cancellation Booking.com reopened the dates in about 19 minutes and Airbnb took 20 to 45.
Then reprice the reopened dates. Their rate was set against demand from weeks earlier and no longer fits the days left (we review rates daily). If the gap is shorter than the minimum stay, that gets fixed too, otherwise it cannot sell at any price. The cleaning slot comes off the schedule and the cash forecast is corrected, since a booking is not money until the payout (from booking to payout).

How should long bookings be thought about?
Cancelled bookings averaged 4.0 nights, against 3.3 for stays that went ahead. The longer the booking, the more nights a single cancellation empties.
A month-long booking empties 28 nights in one go (month-long bookings), and platforms sometimes apply separate rules to long stays, so read the current terms before accepting one.
What should you ask a Japanese management company, and what we will not promise
Ask for cancellation figures before tier names: the share of bookings cancelled on each channel, how far before arrival those cancellations come, and whether a reopened date is repriced the same day (where bookings come from).
27.6% and 58.2% are what our own six rooms in Tokyo and Sapporo recorded. We do not tell an owner their property will match them. What we commit to is measuring the cancellation share and timing of the property we take on over the first months, and choosing its terms again from its own numbers.
A note on method: the reservation records do not say who cancelled, and a change of dates processed as a cancellation and a new booking also counts. The figures are therefore bookings that were cancelled, not guests who changed their minds.
