Booking Holdings told investors on 4 November 2026 that it is rolling out AI tools across its brands to cover the whole trip, from planning to booking to cancelling. On the third-quarter earnings call, CEO Glenn Fogel said Booking.com customers can now cancel bookings and ask questions through its AI, Agoda’s assistant runs in 21 markets from trip ideas through to booking, and Priceline is folding its Penny assistant into the standard experience. The company said customer service costs fell while gross bookings grew 10% year on year, and customer satisfaction stayed high.
The claim, and what actually changed
The excitement will be about the scale. A travel group this size putting AI in front of every customer is a signal that the technology holds up under real volume. Fair enough. But the useful question is the one I always ask first: is there a need, and what changed?
The need is plain. A traveller asks which hotel suits a family of four, books it, and two weeks later comes back to say the plans changed and night two has to go. Same person, same brand, same chat window. Nobody at the traveller’s end thinks of the first message as “sales” and the second as “support”.
What changed is that the AI now covers that whole span and can act on it. Cancelling a booking is an action in a system, not an answer. Booking Holdings reports that this cut service costs while the business grew, and that satisfaction held. That is the part worth noticing.
What did not change is the customer. Travellers have always asked these questions in one place. The company has stopped routing them to separate systems, which is a decision about posture, not a feature.
The customer never split the conversation
This is the position I hold for every business I work with, and it is the one Booking Holdings has just applied at scale. Customer service and sales are one conversation. “Where’s my order” and “which one should I buy” arrive in the same WhatsApp thread.
Look at what actually happens in a shop. The customer asking about a late delivery is the same customer who, three messages later, asks whether the larger size is in stock. A patient messaging a clinic to move an appointment also asks whether the new treatment is suitable for them. A marketing team running a promotion gets “does the discount apply to my existing order” and “can I add one more” in the same breath.
Now picture the operator who bought a sales bot from one vendor and a support bot from another. The handover between them is where the conversation breaks. The sales bot does not know the order. The support bot cannot recommend anything. The customer repeats themselves, or gives up. The operator pays two bills, maintains two sets of rules, and reads two dashboards that cannot see each other.
That split also hides the number that matters. One dashboard counts tickets deflected. The other counts conversions. Nobody counts the order that was saved because a delivery complaint ended in a recommendation and an offer. Booking Holdings reported both lower costs and higher bookings from the same rollout. A split setup cannot even see that result, let alone produce it.
What I build on
AeroChat is built on this assumption. You describe your customer service and sales workflows in your own words, and the same system runs both, connected to your orders, your payments and your delivery. So it can check an order, recommend a product, apply a discount within the limit you set, or send a payment link, all inside the one chat the customer already opened. The limits on refunds, discounts and promises are enforced by the system, so the operator decides what it can commit to.
That matters most at the moments a separate bot would miss. A recommendation at 11pm when the shop is closed. A deposit taken before the customer goes quiet.
What a business should do
Treat the Booking Holdings news as a prompt to look at your own conversations, not as a reason to buy more software.
- Read a week of your chats. Count how many switch between a sales question and a service question in the same thread. In my experience the answer surprises operators.
- If you are evaluating tools, evaluate one system that does both and connects to your order data and payments. Two bots with a handover between them is a worse customer experience than one that acts.
- Put your limits in the system before you switch it on: refund ceiling, discount range, delivery dates you will promise, and when a person takes over with the full context.
- Measure both sides from day one. Conversations handled without a person, and orders from shoppers who talked to the AI first.
Source: CX Dive: Booking Holdings credits AI for better booking, customer service