Ryanair Replaced 12 AI Agents With One. 80% of Chats End There

In February, Ryanair did the opposite of what most operators do with AI. It had twelve specialised customer-service agents — one for baggage, one for refunds, one for bookings, one for disruption — and it deleted eleven of them. Everything now runs through a single agent that handles every category in one conversation.

The consolidated system answers 120,000 chats a day in seven languages and resolves 80% of them without a person, at 94% accuracy. That is not a story about buying more AI. It is a story about wiring up the AI you already have.

What subtraction actually bought

The numbers moved in the direction you would expect if the problem was never the model. Response time fell from 18 seconds to 2.9 seconds — an 84% improvement. Ryanair and its partner Cation Consulting benchmarked five foundation models against 12,000 real production questions before picking one, and got a 25% accuracy uplift out of the switch. Since the first version launched in October 2024, the system has produced 10 million answers and saved more than 500,000 agent hours, and customer-service contacts per passenger carried have dropped 70%.

Tracy Kennedy, Ryanair’s Director of Customer Services, framed the goal in a way that has nothing to do with technology: “We needed to move from linear cost scaling to a model where customer service costs remain largely fixed as passenger volumes grow.” That is the same sentence a venue operations manager says every spring, right before the school-holiday phones start ringing.

Everyone has AI now. Almost nobody has it joined up

New Talkdesk research presented on September 14 puts the gap in numbers: 98% of companies have deployed AI somewhere in the customer journey, but only 15% combine agentic AI with orchestration across departments. The top blockers are compliance (50%), security (48%) and disconnected systems (45%). Talkdesk’s Pedro Andrade put it plainly: “Adoption is easy. The orchestration is the hardest part.”

That is precisely the shape of the problem at an attraction. A visitor’s question is almost never about one thing. Can I get a wheelchair to the 2pm show, and is there parking near that gate, and does my group rate still apply if two people drop out? Three systems, three staff members, one visitor who wanted one answer. Twelve narrow bots would have failed that question too. A single agent that knows the whole venue the way a great front-of-house manager does answers it in one pass.

Where to start before your next peak weekend

Three things, in order, and none of them require buying anything first.

Count your real containment rate. Take last week: of every call, message and DM that arrived, what share got a complete answer without a staff member touching it? Most venues have never measured this. Ryanair’s 80% is the benchmark that matters — not response speed, which is the metric everyone reports because it is the easy one.

Find the cross-system questions. Ask your front desk for the ten things they get asked most, then mark which ones need information from two or more places — ticketing, show schedules, accessibility, parking, group pricing. Those are the ones a single-purpose FAQ widget will never close.

Then join it up rather than adding another tool. One agent covering the website, the phone line and messaging together, in the visitor’s language, beats a chat box on one page and voicemail on the other. The cost of leaving it is quiet: the group booking that came in at 9pm Sunday and never got a reply, and the international family who gave up at the language barrier and spent their day somewhere else.