ImplementationAdoption storyFor: Customer support

Airbnb says AI resolves 40% of support issues; the cost line still rose 8%

Airbnb says AI cut support cost per booking 17% in Q3. Its 10-Q shows outside support spend fell $26M while the whole support line grew to $393M.

At a glance

Company
Airbnb · Technology and software · Size not stated
Team and use case
Customer support · Ticket deflection
Tools
Airbnb's own AI assistant (vendor not named)
Stage
In production, one team
Results
  • Share of all support issues resolved without a human agent: A third (Q2 2026) → About 40% (Q3 2026)
  • Customer support cost per booking, year over year: Down more than 17%
  • Professional and third-party service costs in operations and support, Q3: Down $26 million
  • Operations and support expense, Q3: $365 million (2025) → $393 million (2026)
Cost
Not disclosed. Neither filing gives the cost of building or running the assistant.
People
Spending fell in a category that mainly pays outside support partners; Airbnb's own payroll in the line rose $21 million on higher average headcount. Jobs affected and who reviews the assistant: Not disclosed.
Evidence
Company claim, no method

The company states a result without saying how it was measured.

What this leaves out: The 40% and 17% come from a shareholder letter that defines neither "resolved" nor "cost per booking". The $26 million is from an unaudited 10-Q with a year-earlier baseline. No cost for the assistant and no satisfaction score is given.

Airbnb told shareholders on 5 November 2026 that about 40% of all its customer support issues in the third quarter were resolved without a human agent, up from a third the quarter before, and that customer support cost per booking fell more than 17% in a year, "driven in large part by improvements to our AI assistant". Both figures are in its third-quarter shareholder letter, filed with the US Securities and Exchange Commission (SEC).

The quarterly report (Form 10-Q) filed the same day shows where the saving landed and what it did not offset. The expense line that holds support, "operations and support", rose 8%, from $365 million to $393 million.

What the 10-Q shows

  • Outside support cost fell. Professional and third-party service costs in that line dropped $26 million in the quarter and $42 million over nine months, "due to lower agent contact volume resulting from increased use of artificial intelligence ("AI") in community support". A footnote says this category primarily covers customer support partners, consultants, contingent workers and professional fees.
  • In-house payroll rose. Payroll-related expenses in the same line went up $21 million in the quarter and $61 million over nine months, "driven by higher average headcount".
  • Payouts rose. Customer relations costs went up $33 million in the quarter, "driven by higher make-good payouts and related case reserves". The filing does not connect this to the AI assistant in either direction.
  • As a share of revenue the line fell from 9% to 8%, because revenue grew 19%.

The 17% has no definition

The letter does not say how "customer support cost per booking" is calculated and gives no dollar figure for it. The nearest check the 10-Q allows is rougher: operations and support expense divided by nights and seats booked. That is $365 million ÷ 134 million = $2.72 a year ago and $393 million ÷ 155 million = $2.54 now, a fall of 7%. That line includes customer payouts and other operations work, and a night is not a booking, so it is not Airbnb's measure. It does show that a per-booking percentage and a budget line can move in opposite directions.

"Resolved without a human agent" is not defined either. The letter says the 40% covers all support issues, "not just those that begin in the AI assistant", which is the stricter denominator. It does not say whether a customer who gave up counts as resolved.

The people side

The filing ties the saving to lower contact volume for agents, in a cost category that mostly pays outside partners, while Airbnb's own payroll in the line grew. Neither document says how many partner agents or seats went, who reviews the assistant's answers, or how a customer reaches a person. On the earnings call, CX Dive reports, chief executive Brian Chesky said "customer service NPS is as high as it's ever been", a customer satisfaction score; he gave no number.

Neither document gives the cost of building or running the assistant, so a net saving cannot be worked out. The assistant launched about a year and a half ago, according to CX Dive; the letter says it now works in more than 50 languages and is being extended to phone support in the US.

What to do

  • If you run support, ask finance for the three lines Airbnb's filing separates (partner or outsourcer spend, in-house support payroll, and refunds or goodwill payouts) for the quarter before your assistant went live and the latest one, in dollars.
  • When a vendor or your own team quotes a share of issues resolved by AI, ask whether the denominator is every issue, as Airbnb's is, or only conversations the assistant touched, and what happens to abandoned ones.
  • If the saving is meant to come from an outsourcing contract, have procurement check its volume minimums before the forecast goes to the CFO.
Was this useful for your work?

Take this to work

Forward this

Airbnb says AI resolved about 40% of its support issues last quarter and cut support cost per booking more than 17%, yet its SEC filing shows the expense line that holds support rose 8% to $393 million as payroll and customer payouts grew. Before we report a per-ticket saving, we should show partner spend, in-house payroll and payouts together for the same periods.

Questions to ask

  1. Is our share of issues resolved by AI counted against every issue or only those the assistant touched, and do abandoned conversations count as resolved?
  2. What did partner spend, in-house support payroll, and refunds or goodwill payouts each do between the quarter before launch and the latest quarter?
  3. What does the assistant cost to run and maintain a quarter, and is that inside the cost-per-ticket figure we report?
  4. What volume minimums are in our outsourcing contract, and when can they be renegotiated?
  5. Who reviews the assistant's answers, and how does a customer reach a person?

The arithmetic

Support cost per booking = (partner spend + in-house support payroll + refunds and goodwill payouts + assistant running cost) ÷ bookings, for the same quarter this year and last.

Airbnb gives only the result (down more than 17%) and none of the inputs. The nearest check from its 10-Q: operations and support expense ÷ nights and seats booked = $365M ÷ 134M = $2.72 in Q3 2025 and $393M ÷ 155M = $2.54 in Q3 2026, down 7%. That line holds more than support, and the assistant's running cost is not stated.

Questions this article answers

How much of Airbnb's customer support does AI handle?

Airbnb's third-quarter 2026 shareholder letter says about 40% of all customer support issues were resolved without a human agent, up from a third in the second quarter. The letter does not define "resolved".

How much has AI saved Airbnb on customer support?

Airbnb says customer support cost per booking fell more than 17% year over year in the third quarter of 2026, without giving a dollar figure. Its 10-Q reports a $26 million fall in professional and third-party service costs in the quarter, attributed to AI lowering agent contact volume.

Did Airbnb's total support spending go down after it deployed AI?

No. Operations and support expense rose 8%, from $365 million to $393 million, in the third quarter of 2026, because higher payroll and higher customer payouts outweighed the saving on outside services. It fell from 9% to 8% of revenue.

Did Airbnb cut support jobs because of its AI assistant?

Neither the shareholder letter nor the 10-Q gives a number of jobs. The 10-Q reports lower spending in a category that mainly covers outside support partners and higher payroll from higher average headcount at Airbnb itself.

Send this to someone who runs a teamLinkedInXBlueskyHNEmail