Airbnb's AI support agent resolves about a third of issues and leans on Alibaba's Qwen

The problem

Airbnb serves guests and hosts in several languages and markets, and every support contact that reaches a person adds to its customer service cost base. The company wanted more issues resolved without a human agent.

What they built

Airbnb built its own AI customer service agent. Users message an AI assistant, and the issue is resolved without a human agent when the assistant can handle it. By Q2 2025 the agent served all US users. By the Q4 2025 results release it covered English, French and Spanish speakers in the US, Canada and Mexico.

CEO Brian Chesky told Bloomberg that Airbnb relies heavily on Alibaba's Qwen models. Qwen is an open-weight model family, meaning its files are published so a company can run and adapt them. Chesky described Qwen as "fast and cheap." He also said ChatGPT's integration was not quite ready for Airbnb's needs. A report relayed by MENA Startup Digest says Airbnb uses 13 models in total, including ones from OpenAI and Google.

How they did it

The sources describe a staged rollout rather than a single launch. The agent reached all US users in Q2 2025, then moved to Canada, Mexico and more languages. Airbnb says it plans a global rollout later in 2026.

The sources do not give a budget, team size or build timeline. They also do not say which Qwen version Airbnb uses or where it runs.

Results

In Q2 2025 Airbnb said the agent lowered the share of hosts and guests who had to contact a person by 15%. In its Q4 2025 results release it reported that about a third of issues were resolved without needing an agent, with a significant drop in resolution time. TechCrunch repeated the one-third figure from the earnings call, where Chesky said the cost base of customer service would fall and quality would rise.

Airbnb aims for AI chat and voice to handle more than 30% of all tickets within a year, in every language where it has human agents.

What went wrong

In April 2026 two US House committees wrote to Airbnb about its use of Qwen. They cited national security and data security concerns and asked which Chinese models Airbnb uses and how data reaches them. The letter is an inquiry. It does not report a violation, and Airbnb's reply is not in the sources.

What to copy

Report two numbers on a regular basis: the share of issues the agent closes alone and the change in contacts reaching people. Expand one market and language at a time. Review a model's origin and hosting with your legal and security teams before you commit to it.