We’ve automated a lot of our guest messaging. Door codes, directions from the station, luggage storage, trash separation, how the air conditioner works: a chatbot grounded in each property’s own information answers those in seconds. In our experience, the bot handles routine property questions quickly.

One group of messages still comes to me, though, and I don’t think that’s going to change: “Can we move our dates?”, “One of us can’t come anymore, can we get money back?”, “We arrived late because of a typhoon, can you refund the first night?” These are the ones that take real judgment, so this post covers how we handle them.

TL;DR

  • Reservation changes and refund requests are the main guest messages that still need a human, because each answer costs or earns you money.
  • Record changes in the booking platform and follow its approved payment and refund process. Where the property collected payment under an authorized arrangement, process the refund through that payment channel and keep the receipt and platform correspondence. Diverting payment without authorization can breach the platform’s terms or limit your protection.
  • A change in dates, guest count or effective nightly rate can change the accommodation tax you owe, and in jurisdictions with price bands a discount can move a stay into a different band.
  • First check whether a platform policy requires a cancellation, refund or rebooking support. After that, a simple four-question check (calendar, sunk cost, policy, goodwill) makes most discretionary decisions quick and consistent.
  • Reply fast even when the answer is “let me check.” A prompt acknowledgment calms most guests down.

Why can’t a chatbot handle reservation changes and refunds?

We don’t let a chatbot handle them because every answer is a financial decision, and the right one depends on context we want a person to look at. The bot can explain your cancellation policy. But we require a person to check live availability, incurred costs and guest history before approving a refund or change: whether the week around the new dates is already fully booked, whether the cleaner has been paid, whether this guest has stayed with you three times before.

Most guest questions have answers that are specific to the property but don’t change. Refund and change requests are the opposite. They’re about one reservation, they’re often emotional, and whatever you say can be quoted back to you in a review or a platform dispute. So we let the bot recognize them and pass them to a person, and it’s designed not to promise anything.

What kinds of change requests do we actually see?

Almost all of them fall into five types:

  1. Date shifts: same number of nights, different dates. These are often easy if the new dates are open.
  2. Shortened stays: a flight changed or plans fell through, and the guest wants money back for the nights they won’t use.
  3. Guest-count changes: someone drops out or joins. This can affect pricing, bedding and, as I explain below, tax.
  4. Late arrivals or early departures because of disruption: typhoons, cancelled shinkansen, missed connections. In a Japanese autumn this is very real.
  5. Complaint-driven refunds: “the room wasn’t what we expected.” This is really a service-recovery conversation that looks like a refund request.

Sorting a message into one of these first stops you from treating every request as a crisis. A date shift and a complaint-driven refund need completely different responses.

Should you handle changes on the platform or off it?

Record changes in the booking platform where the reservation was made and follow its approved payment and refund process, even when a guest suggests “just sending the difference by bank transfer.” Platforms differ here. Airbnb, for example, sets out its payment requirements, how reservation alterations work and the limited exceptions it permits in Pricing and payment transparency. Booking.com lets properties collect payment from guests directly under its arrangements (Booking.com: How we work, sections 1H and 1L). Where the property collected payment under an authorized arrangement like that, process the refund through that payment channel and keep the receipt and the platform correspondence.

There are three reasons for this. First, diverting payment off-platform without authorization can breach the platform’s terms or limit the host protection and dispute support you would otherwise have. Second, your records need to match your payouts. At tax time, a refund that exists only in a chat thread is a reconciliation headache. Third, it isn’t worth risking your listing to save a few minutes.

For direct bookings, the same principle applies to your own system. Record the change in the reservation itself, not just in the conversation.

What happens to accommodation tax when a booking changes?

A changed booking can change the accommodation tax you owe, so update the numbers along with the reservation. Japan’s accommodation tax is set by local jurisdictions such as prefectures, cities and towns, and is usually charged per person per night. Changing nights usually changes the tax proportionally, but a change in guest count should be recalculated from the final per-person, per-night accommodation charge. In banded systems, fewer guests can sometimes increase the per-person rate. For example, under Tokyo’s rules through March 31, 2027, a hotel/ryokan twin room with a tax-exclusive accommodation charge of ¥20,000 per night incurs ¥200 for one occupant or ¥100 each for two occupants (東京都「宿泊税の手引」).

The less obvious issue is the rate. Jurisdictions such as Tokyo Metropolis and Osaka Prefecture use price bands, where the amount per person depends on the per-person nightly accommodation charge. If the operator itself discounts or refunds the accommodation charge, the discounted amount may be the tax base, and in some cases that moves the stay into a different band. Third-party points, subsidies or OTA-funded discounts may be treated differently, so check the local FAQ. Deciding whether the band is based on the rate before or after cleaning fees is already a common filing mistake we see, and a mid-stay adjustment makes it easier to get wrong. For example, Tokyo and Osaka include mandatory cleaning fees in the accommodation charge, while meals and consumption tax are excluded (東京都主税局「申告書作成 チェックポイント」; 大阪府「回答(宿泊税)『制度の概要に関すること』」). Always confirm the local rule.

Tokyo is also changing its rules. Its current ¥100/¥200 bands apply to hotels and ryokan until 2027-03-31. From 2027-04-01, Tokyo will also cover simple lodgings and minpaku, with a 3% tax on stays of ¥13,000 or more per person per night (東京都主税局「宿泊税の見直し」).

Don’t assume the OTA payout has handled accommodation tax. Check the reservation and payout records and your local rules. In many jurisdictions the accommodation operator is the special collection obligor and must reconcile and remit the tax, so you’re the one who has to catch this. We recalculate with the final reservation data, and this is one reason we built japan-stay-tax as an open-source library: the calculation should come from the booking data, not from memory. Also keep your guest register accurate when the names or number of guests change; under the residential accommodation business (minpaku) framework, the register must accurately record every guest (観光庁「事業者の業務[1]」).

How do we decide whether to say yes?

First we check whether an applicable platform policy requires cancellation, refund or rebooking support, and we apply our own goodwill framework only to discretionary concessions after that. On Airbnb, for example, the Major Disruptive Events Policy can override your cancellation terms for qualifying events at the destination, although ordinary transport disruption and foreseeable seasonal weather generally don’t qualify by themselves. Its Rebooking and refund policy for homes can require relief for covered problems with the accommodation; guests generally must report them within 72 hours of discovery.

After that, we ask four questions in this order:

  1. Calendar: Can we realistically resell the freed nights? A shortened stay in peak koyo season is very different from one on a quiet Tuesday in February.
  2. Sunk cost: What have we already spent? Cleaning scheduled, linens prepared, consumables stocked.
  3. Policy: What did the guest agree to? We start from the published cancellation policy, and it gives us a clear default.
  4. Goodwill: Would a small concession protect a review, a repeat guest or our own sense of fairness? Weather disruptions that no platform policy covers often land here.

When the calendar is flexible and our costs are low, we usually agree to date shifts. For shortened stays we often offer a partial refund that depends on whether we manage to rebook the nights. Complaint-driven refunds get a fix first (a cleaner sent back, an item replaced) and money second, unless a platform policy requires a refund or rebooking for the problem.

What should the reply look like?

Reply quickly, even if you don’t have the answer yet. Something like: “Thanks for letting us know, and sorry about the flight. I’m checking the calendar now and will confirm the options within a few hours.” That takes the pressure off. Then give a clear, specific offer, point to the platform’s alteration request, and don’t use vague phrases like “we’ll see what we can do.”

Consistency is what keeps this manageable. Write down how you’ve handled past cases so that the next person who covers your inbox, or you six months from now, can make the same call.

FAQ

Q: Should I give refunds outside Airbnb or other OTAs to save time?

Not as a shortcut. Record changes in the booking platform and follow its approved payment and refund process. Where the property collected payment under an authorized arrangement, process the refund through that payment channel and keep the receipt and platform correspondence. Diverting payment without authorization can breach the platform’s terms, limit dispute support and cause problems when you reconcile your payouts.

Q: Do I have to recalculate accommodation tax if a guest shortens their stay?

Generally yes. Fewer nights usually reduces the tax proportionally, and a change in guest count should be recalculated from the final per-person, per-night accommodation charge. In jurisdictions with price bands, an operator-funded partial refund can also change which band applies. Check your local jurisdiction’s rules.

Q: Can a chatbot help with refund requests at all?

Yes, as a triage step. It can recognize the request, collect the reservation details, explain the published policy and hand the conversation to a human. It shouldn’t commit to any refund or change on its own.

This post is for informational purposes only and does not constitute legal or tax advice. Please consult a qualified professional for your specific situation.