About 40% of Airbnb's customer support problems are now closed without a human agent, and the cost of support for each booking has fallen more than 17% from a year ago. The assistant is also being extended to phone support in the US, so a guest locked out at night can talk to it instead of typing.
The machine takes the short cases first
If every case cost the same, closing 40% of them with no person would cut the support bill by 40%. The bill fell by 17%. Part of the gap is that the assistant was already doing some of the work a year ago, when it had cut the number of people who needed a human agent by around 15%, so 17% is what the latest step added. The other part is that the machine takes the short cases first: where is my booking, change my dates, send me the receipt. What reaches a person is a host who cancels the night before, or an argument over damage, and those take longer and cost more per case.
So the share of cases a machine closes overstates the saving, and it also describes the human work badly. A support agent at a company like this will spend the day on fewer, longer and more serious problems.
Where the saved money goes
The bank branch shows what usually happens next. When running a branch got cheaper, banks opened more branches and the tellers moved toward selling and relationship work. Support has its own version of opening more branches: serving more people in more ways. Airbnb's assistant already works in more than 50 languages, and phone support is a channel that once needed a person on every call. A company that could not afford a phone line in a small market can now afford one, and every extra channel produces some hard cases that need a human.
The phone is also where the urgent problems live. A guest standing outside a locked door does not want a menu. The voice assistant takes the first minute, and if it cannot solve the problem, it has to pass it on with the story already written down. When the guest has to start again with a person, the company has saved money by spending the customer's patience.
What a small business can take from it
Take a shop that ships 1,000 orders a month and gets a message about one order in seven, so around 140 messages. If an assistant closes 40% of them, that is 56 messages, about five hours a month at five minutes each. That is a small saving for the owner. The real change is that those 56 customers get an answer at 3 in the morning instead of the next day, and the 84 messages left are the lost parcels and refund disputes that deserve the owner's full attention.
That is why the useful measure is the cost of support per order, together with how fast the hard cases get resolved, and not how many cases the machine closes. When you choose or judge any support system, including your own, ask what happens to the problem the assistant cannot close: whether it reaches a person who can decide, with the whole conversation in front of them.