Ryanair
Ryanair runs a customer chat agent on Amazon Nova that resolves 80% of chats without staff
Ryanair runs a generative AI chat agent for customer service on Amazon Bedrock, in seven languages. AWS reports that 80% of about 120,000 daily chats are contained, accuracy of 94% since the October 2024 launch and replies in 2.9 seconds instead of 18.
Key figures
Copied exactly from the sources. Copy any figure with its citation.
80%
Containment rate of the chat agent (chats handled without a human agent)1
120,000
Daily chat interactions1
94%
Accuracy rate (AWS page; the same page's quote calls it a success rate)1
10 million
Chats answered since the October 2024 launch, at the 94% accuracy rate (PYMNTS)2
What it took
The problem
Ryanair carries a very large number of passengers, and customer questions about baggage, refunds, bookings and disruptions grow with them. Handling those contacts with staff alone would grow costs with passenger volume.
What they built
A generative AI chat agent that answers customer questions in seven languages. It runs on Amazon Bedrock AgentCore and uses Amazon Nova 2 Lite. Six languages besides English go through Amazon Translate. In February 2026 Ryanair merged 12 separate agents for different topics into a single agent.
How they did it
Ryanair worked with AWS and the AWS partner Cation Consulting. It tested five models on 12,000 real production questions and picked Nova 2 Lite. The chat service launched in October 2024.
Results
AWS reports a containment rate of 80% on about 120,000 daily chats, accuracy of 94%, and a reply time cut from 18 seconds to 2.9 seconds. PYMNTS adds that the system has answered 10 million chats since launch. These are vendor-published figures with no method given.
All reported figures
| Measure | Figure | Citation |
|---|---|---|
| Containment rate of the chat agent (chats handled without a human agent) | 80%1 | |
| Daily chat interactions | 120,0001 | |
| Accuracy rate (AWS page; the same page's quote calls it a success rate) | 94%1 | |
| Chats answered since the October 2024 launch, at the 94% accuracy rate (PYMNTS) | 10 million2 | |
| Reply time | from 18 seconds to 2.9 seconds1 |
Lessons for your project
Test models on your own questions. Ryanair compared five models on 12,000 real customer questions before choosing one.
Merge specialist bots once the base is stable. In February 2026 Ryanair combined 12 domain agents into one agent.
Report containment and accuracy separately. A bot can keep chats away from staff and still answer wrongly.
Keep voice and chat results apart. AWS gives a 500,000 agent-hour saving for the combined phone and chat systems, so it should not be credited to chat alone.
Caveats
- The figures come from an AWS case study, so the main source is the vendor. PYMNTS, dated 15 September 2026, relays the same figures and is not independent confirmation.
- The AWS page shows no publication date. Its only date is a copyright footer, which is not used.
- AWS labels the 94% figure an accuracy rate in one place and a success rate in a pull quote. The method is not stated.
- The 500,000 agent hours and the 70% fall in agents per passenger carried are given for the combined voice and chat systems, so they are not listed as chat results. A later AWS sentence words the 70% as a fall in customer service contacts per passenger carried by the assistant; the page is inconsistent, so it is not used.
Sources
- Ryanair agentic AI
Amazon Web Services · Undated · Vendor case study · aws.amazon.com
- Ryanair's AI Resolves 80% of Customer Chats Without Humans
PYMNTS · 2026-09-15 · News · pymnts.com
Summarized in our own words from the sources above. Spotted an error? See corrections.