Airbnb proves AI cuts support costs by 16%: what that means if you don't have their numbers

For years the corporate narrative around artificial intelligence has rested on soft metrics: employees trained, chatbots launched, pilot projects kicked off. This earnings season, according to a PYMNTS analysis, has finally delivered something more concrete — and two examples in particular are worth paying attention to. Monday.com told investors that annual recurring revenue tied to its AI products doubled between Q1 and Q2 2026, reaching 17% of the company's net new ARR: customers are actively choosing AI features and willing to pay extra for them. Airbnb showed the other side of the same coin: the customer support cost per booking dropped roughly 16% year-on-year, thanks to an AI assistant that resolves nearly 45% of guest issues without any human intervention, while AI cut the time between idea and feature launch by up to 60%, allowing the company to ship nearly 80% more features in the first half of 2026 compared to the previous year.

Why these numbers matter more than they seem

The central point of the PYMNTS analysis isn't any single figure — it's the fact that boards have stopped settling for adoption metrics (how many people are using a given tool, how many pilots have been launched) and are now demanding numbers tied directly to revenue or costs. This is a trend confirmed on a broader scale by PYMNTS Intelligence, which has been monitoring corporate sentiment on generative AI since March 2024 across more than a thousand observations of companies with at least a billion dollars in revenue: the share of executives reporting a favourable outcome has risen to 96%, even though most of those same executives admit the technology is still in an early phase.

There is, however, a detail that makes this enthusiasm less unanimous than it sounds: according to Gartner data cited in the same article, 45% of CFOs are still directing their AI budgets towards generic productivity gains rather than measurable strategic outcomes, while global spending on AI platforms and models is forecast to reach $64.25 billion in 2026 — up 63.4% from 2025 — with total AI spending, infrastructure included, estimated by Gartner at $2.52 trillion for the year. Wedbush analysts, quoted in the same piece, note that many companies have run AI pilots without any framework for measuring success, and were subsequently unable to justify the investment even after significant spend. This is exactly the problem I had already tackled when discussing the vendor-evaluation framework proposed by an enterprise technology executive, and finding it confirmed by an analysis this different — one oriented squarely towards corporate finance — tells me it's a real, widespread problem, not an isolated observation.

The limits of these two examples, honestly stated

Monday.com and Airbnb can demonstrate these results because they operate at enormous scale: millions of transactions, millions of bookings, which make it possible to calculate with statistical precision a before and an after. An independent hospitality property will never be able to produce a number like "16% less in support costs per booking," simply because its own volume of requests is too small for a statistically meaningful comparison. It would be dishonest to imply that the same rigour is reproducible at any scale.

What still holds, even in small

The underlying principle, though, is perfectly applicable: before introducing an AI tool, decide clearly what you want to measure — and actually measure it, before and after. You don't need a finance department for that. In my own property, for example, before introducing any new tool, I track a few simple indicators I can follow on my own: the average response time to guest requests, the number of requests I handle personally versus those managed autonomously by a tool, the time I spend each week on repetitive tasks like replying to reviews. These aren't numbers to present to a board, but they are exactly the same principle behind Airbnb's and Monday.com's declarations: one number before, one tool, one number after — and only then an informed decision on whether to keep going, adjust, or drop that tool. Even at the tiniest scale, that's the difference between a technology investment made with method and one made out of fashion.

FAQ

Q: How did Airbnb measure the return on its AI investment? A: Airbnb told investors that the customer support cost per booking dropped roughly 16% year-on-year, thanks to an AI assistant that resolves nearly 45% of guest issues without human intervention.

Q: Why is it difficult for a small hospitality property to measure AI ROI the way Airbnb or Monday.com do? A: Because these companies operate on enormous transaction volumes that allow a statistically significant before-and-after comparison, whereas a small property has volumes too low for the same type of measurement.

Q: What can a small property do to measure the real impact of an AI tool? A: It can identify a few simple indicators to track on its own — such as average response time to guests or time spent on repetitive tasks — measure them before and after introducing the tool, and decide accordingly.

*cover image generated with AI