The Dispute Resolution Landscape

In the OTA era (2000–present) the party that decides a guest–hotel dispute is almost never a neutral one with knowledge of the property. It is a platform support desk whose commercial customer is the traveler, or a card issuer whose contractual customer is the cardholder. That is not a moral failing; it is a fact about whom those institutions serve. OTAs built the demand that fills independent hotels and still deliver it. The objection here is narrower: the gatekeeper position also makes them the judge, and the judge has a side.

This is collaborative research. It describes four dispute channels as they stand in September 2026, is plain about the limits of public data, and then sets out the model the Dispute Resolution Working Group (DR-WG) is drafting as ADAPT-DRP.

Booking.com manages its property relationships through the Partner Hub and the extranet. Most reservations run on the agency model: the guest pays the property, and Booking.com invoices commission after the month of checkout. Under Payments by Booking.com, the platform collects from the guest and pays the property after the stay, typically by virtual credit card or bank transfer. That split decides who holds the money when the argument starts. On agency reservations the property usually has the funds, so a complaint becomes a request from customer service for a refund or goodwill gesture, which the property can decline. On platform-collected reservations Booking.com can refund the guest and net the amount against the payout. There is no formal arbitration step and no published service level for partner-side complaints.

The property's tools are narrower than the guest's. The extranet lets a property mark a no-show within 48 hours of the scheduled arrival, which removes the commission unless the property charges a no-show fee, and report guest misconduct — damage, unacceptable behavior, unpaid charges — within a short window after checkout; the report can affect the guest's account standing but does not by itself move money. When the property is at fault the terms are explicit: a property that cannot honor a confirmed reservation must relocate the guest to comparable or better accommodation and bear the price difference plus reasonable transport.

Expedia Group runs the lodging relationship through Expedia Partner Central. Under Expedia Collect, Expedia is the merchant of record: it charges the traveler at booking and pays the property after checkout, historically by virtual card or direct deposit. Under Hotel Collect, the property charges the guest and Expedia invoices commission. Refund flows follow the money: on Expedia Collect reservations a refund is Expedia's to grant, and the property's next payout is adjusted; on Hotel Collect reservations Expedia support can request a refund but cannot execute one. Partner-side disagreements — an invoice that includes a cancelled stay, a payout that does not match the folio — are raised as cases in Partner Central, resolved in days to weeks, with no published partner SLA.

Relocation is again the clearest rule in the book. If a property cannot honor a confirmed booking, Expedia relocates the traveler to a comparable or better property and bills the original property for the rate difference and transportation. Both large OTAs handle the situations where the hotel is clearly wrong with predetermined remedies, and everything ambiguous through a support queue whose commercial priority is the traveler's next booking.

Airbnb is the most instructive case, because its dispute system has visibly moved with its supply position. The Resolution Center is a negotiation tool: either party can request or send money, the other has 72 hours to respond, and if they cannot agree either side can ask Airbnb to step in. AirCover for Hosts, introduced in November 2021 and expanded in November 2022, provides $3M in host damage protection and $1M in liability coverage, with damage claims filed within 14 days of checkout.

The history explains the sentiment. In 2011, after a widely reported ransacking of a host's home, Airbnb introduced a $50,000 Host Guarantee, raised to $1M the following year; host liability insurance followed in 2015. Supply was the constraint, and policy protected supply. The swing came in March 2020, when Airbnb refunded pandemic cancellations in full across hosts' own policies; on March 30, 2020, after loud host objections, it committed $250M to pay hosts 25% of what those policies would have yielded — an acknowledgement, not a reversal.

What followed reads as an attempt to hold both sides at once. AirCover for guests (May 2022) added a get-what-you-booked guarantee, and the Rebooking and Refund Policy gave guests 72 hours after discovering a problem to seek a refund or rebooking. In June 2024 the Major Disruptive Events Policy replaced extenuating circumstances with a narrower list of covered events, a modest step back toward host terms. Then, beginning in October 2025 for PMS-connected listings and completing on April 13, 2026, Airbnb moved software-connected hosts to a single 15.5% host-only fee, a change professional hosts read as a tax on professionalism.

The pendulum reading is this. When hosts were scarce, the rules protected hosts. When hosts became abundant, the rules protected the party that generates the next booking, and host protections were rebuilt afterward as insurance rather than adjudication rights. The cost is a cohort of experienced hosts who now warn others against listing. None of this requires bad faith; it is what a two-sided platform does when one side stops being the constraint.

The card networks are the dispute channel of last resort for every direct booking and the default one for prepaid, non-refundable rates. The process was built for unauthorized transactions and adapted, imperfectly, to service disputes. A cardholder may dispute most consumer transactions within 120 calendar days of the processing date or, for services not yet delivered, of the expected service date, with an outer limit of 540 days. Under Visa's post-2018 framework, hospitality cases cluster under reason codes 13.1 (services not received), 13.3 (not as described or defective), 13.7 (cancelled services), and 10.4 (card-absent fraud); Mastercard groups most of the same situations under 4853 (cardholder dispute) and 4837 (no cardholder authorization).

The issuer typically credits the cardholder provisionally when the dispute opens. The hotel's acquirer sets a deadline, often measured in days, to submit representment — registration card, accepted rate rules, folio, correspondence — and the issuer reverses or upholds the chargeback. If the parties still disagree, the case can move to pre-arbitration and then network arbitration, where filing fees run to several hundred dollars and the losing side pays.

Visa Compelling Evidence 3.0, effective April 2023, is often cited as relief for merchants. It is real but narrow: it lets a merchant defeat or pre-empt a 10.4 fraud dispute by showing two prior undisputed transactions with the same cardholder in the preceding 120–365 days that share at least two data elements, one of which must be a device ID or IP address. It helps a hotel answer 'I never made this booking.' It offers nothing against 'the room was not as described,' which is where most hospitality disputes live.

The costs are asymmetric and mostly fixed. Acquirers charge a per-dispute fee, commonly $15–100, regardless of outcome; the reversed amount is gone until the case is won; and both networks run monitoring programs that fine, and eventually terminate, merchants whose dispute ratio exceeds thresholds around 1% to 1.5% of transactions. The guest pays nothing to open a case.

The commissioned abstract promised data on win rates and financial outcomes. Here is what the public record supports, and what it does not.

Booking.com, Expedia Group, and Airbnb do not publish how often partners prevail in disputes, how long resolutions take, or how much is refunded over partner objections; their published material covers procedures, not outcomes. Card networks publish time limits, reason codes, and monitoring thresholds, not merchant win rates; the percentages that circulate come from dispute-management vendors reporting on their own clients. ADAPT's own draft framework cites operator estimates of how often OTAs side with guests; we treat those as anecdote until the DR-WG's decision database produces measured figures, and would rather publish a table of rules than a number we cannot source.

The table makes two facts plain: in every incumbent channel the decision-maker holds the money or the relationship, usually both, and the guest's cost of opening a dispute is zero while the hotel's is positive whoever is right. ADAPT-DRP, the Dispute Resolution Protocol being drafted by the DR-WG (ADAPT-WG-001), inverts both. The decision-maker is a certified, local arbiter with no stake in either party's next booking, and both parties post something a bad-faith filing can cost them.

Localized certified arbiters

The founding framework, published for committee review in March 2026, opens certification to three kinds of candidates: hospitality domain experts (former general managers, consultants, hotel-school faculty); institutions (tourism boards and DMOs, hotel associations, consumer-protection agencies, ombudsman offices); and legal and ADR professionals. Employees of participating properties and of AI-agent companies are ineligible, as is anyone with a financial relationship with a party. Certification runs through screening, examination, ten co-signed probationary decisions, and annual recertification with random audit, at three levels: Associate, Certified, and Master. Assignment matches jurisdiction and market type, screens for conflicts, balances workload, and randomizes the final pick to prevent forum shopping; either party may challenge an assignment once, free.

Service levels by severity

Guest trust deposits

The guest side of the bond is a small, one-time, refundable trust deposit posted on joining the network, never touched unless a filing is found fraudulent. Guests carry one of three statuses visible to properties and AI agents at booking: Verified (clean history, deposit posted), Bonded (deposit insured by a third party, a service market on the guest side), and New (no history yet, deposit required). Properties can offer verified and bonded guests better rates or waived incidental holds. Pattern controls run both ways: guests whose dispute rate crosses a threshold, and properties whose damage-claim rate does, are flagged.

Settlement is where the model differs most from a chargeback. At booking, the terms — cancellation window, deposit, dispute clause — are attached to the transaction. A dispute reserve, a percentage of the booking value, stays in escrow until checkout plus a 48-hour filing window; the rest settles on the encoded schedule. If a dispute is filed, only the contested amount is frozen, and the arbiter's written decision is itself the settlement instruction: release, refund, or split. Arbiters are paid a fixed fee by tier ($75–150, plus a per-day supplement on complex cases) from a network fee of 0.5–1.0% of booking value, so compensation is independent of outcome and amount. Tier 1 cases can trigger an emergency release for alternative accommodation before the full decision.

1. 1

   Dispute filed

   Either party files through the protocol. The system classifies category and severity, assigns an arbiter within two hours, notifies both parties of the timeline, and freezes only the disputed amount.

2. 2

   Evidence submitted

   Both sides upload within the tier's window: the listing snapshot as it stood at booking, the AI agent's conversation log, the encoded terms, timestamped photos and video, staff and maintenance records. Non-response creates an adverse inference.

3. 3

   Arbiter reviews

   The arbiter applies the framework's standards — the listing is the contract, encoded terms are binding, remedies are proportional, reasoning is evidence-based — and may take a brief written or recorded response from each party. There is no live hearing.

4. 4

   Decision rendered

   A written decision states the dispute, the evidence considered, the reasoning, and the remedy, and is recorded, anonymized, in the network's decision database for consistency and public audit.

5. 5

   Settlement executes

   The decision goes to the settlement layer and the escrow moves accordingly, with no invoice, wire, or collection step. Either party may appeal within 14 days to a three-arbiter panel, one of whom holds legal or ADR certification.

The abstract claims structurally better outcomes, and the word is chosen carefully. The improvement does not depend on arbiters being wiser than support agents. It comes from four changes to the structure of the decision.

• Symmetric cost of a filing. The guest's deposit and the property's reserve mean a frivolous claim costs the claimant something and a legitimate one costs nothing. Incumbent channels price a dispute at zero for one side and a fixed fee for the other.
• Money moves once. Today funds settle and are clawed back weeks later, or are held and released unilaterally. Under escrow the contested amount waits, and the decision is the transfer: no representment, no provisional credit, no reconciliation.
• No downstream customer. Fixed, outcome-neutral fees and randomized assignment remove the incentive that makes a support desk lean toward the traveler and an issuer toward the cardholder.
• Local knowledge is admissible. A Marrakech riad's rooftop terrace, a Kyoto ryokan's shared bath, and a Nashville hotel's Broadway noise are judged by people who know what a reasonable guest in that market expects.

The model has open questions, and the working group has published them: whether a local arbiter can be neutral toward an owner they know; whether guests will trust an arbiter over a brand; what happens where binding consumer arbitration is restricted; whether escrow creates new fraud vectors; whether a trust deposit deters first-time bookers. The inaugural event is a debate with a seat reserved for the opposing case for that reason. The framework also keeps the courts in reach: criminal conduct, personal injury, discrimination, and claims above a proposed $5,000 threshold go to court, and where binding arbitration is unenforceable the arbiter's finding stands as admissible evidence.

What ADAPT proposes. The DR-WG carries this work as ADAPT-WG-001: a dispute taxonomy, the arbiter certification standard (the future ADAPT-WG-004 program), the resolution protocol with tiered SLAs, escrow integration with programmable settlement, the fairness-score system, the guest trust and bonding standard, and the anonymized decision database. Its success metrics are public: over 90% of disputes resolved within SLA, under 10% of decisions overturned on appeal, certified arbiters in the top 50 tourism markets within 18 months, and total dispute cost under half the OTA or chargeback equivalent. The first pilot is planned at the Exchange Building and two or three volunteer properties. Operators, consumer advocates, ADR professionals, and DMOs who want to become or certify arbiters can join the weekly sessions; skeptics are asked for by name.