Airbnb resolves 40% of customer inquiries without human agents, cuts support costs 17%

Airbnb, the home-sharing marketplace, said in earnings materials filed with the U.S. Securities and Exchange Commission (SEC) that about 40% of customer inquiries were resolved without going through a human agent. Customer support costs per booking fell more than 17% from a year earlier. It is a case of a company that uses AI, rather than one that builds it, showing the effect of AI as a figure for how much costs went down.

What it means for Japanese companies: Companies that run call centers or customer support desks, or that outsource them, can use this as a ready-made yardstick. It lets them judge whether to adopt AI by how much support costs per transaction fall, not by how many people they can cut.

What the filing says

There are two figures. One is that roughly 40% of inquiries were handled without a human agent. The other is that customer support costs per booking fell more than 17% from the same period a year earlier.

The document is an exhibit to a filing that Airbnb issued to investors and also submitted to the SEC. Because it is primary information released by the company itself, the figures can be treated as the company's official disclosures. A company is accountable for numbers it puts in its earnings materials, which gives them more weight than claims made in a marketing blog post or a conference talk.

Why cost per booking matters

Until now, discussions of AI's impact have mostly been framed as "AI handled X% of the work" or "it saved Y hours." Those claims don't show how a company's money actually moved. If AI usage fees pile up, the workload can shrink while costs stay the same.

Airbnb's figure divides cost by the number of bookings. That makes it comparable whether bookings rise or fall, and it reflects the outcome after AI usage fees are included. A drop of more than 17% on that basis means support got cheaper even after paying for AI.

Over the past month, the AI conversation has centered on benchmark scores for new models, talent poaching between labs and preparations for IPOs. Meanwhile, in travel booking, an ordinary line of business, AI had begun to show up in the cost figures.

Caveats

The figures were counted by Airbnb itself, and the company decides what counts as "resolved." The number alone cannot tell whether users actually got their problem solved or simply gave up and stopped asking.

Nor is the entire 17%-plus decline necessarily due to AI. Changes to how support is organized, or shifts in the volume of inquiries, may also play a part. The remaining roughly 60% of inquiries are still handled by people. The more complicated cases, such as refund disputes and problems during a stay, are likely to remain with humans.

How Japanese companies could follow suit

First, measure support costs per inquiry, or per transaction, before introducing AI. Without a baseline, the effect cannot be expressed in numbers.

Second, define "resolved without a human" in advance. One option is to count only cases where the same customer does not contact the company again about the same issue within a few days. Companies that outsource their support desks can get there fastest by asking vendors to report these two figures when contracts come up for renewal.

Source: Airbnb's earnings materials filed with the SEC (Exhibit 99.1)