Air India
Air India's AI.g virtual assistant on Azure OpenAI automates most customer queries
Air India launched AI.g in 2023, a generative AI virtual agent built on Azure OpenAI Service that answers passenger questions about bookings, flights and baggage. Microsoft reports 97% of queries fully automated and nearly 4 million queries handled by late 2024.
Key figures
Copied exactly from the sources. Copy any figure with its citation.
97%
Customer queries handled with full automation1
Nearly 4 million
Customer queries handled1
About 30,000
Questions per day on average1
About 1,300
Topic areas covered1
What it took
Cost, team size and timeline were not disclosed in the sources.
The problem
Air India's passenger numbers have doubled since early 2022, according to Microsoft's case study, and the airline wanted to handle that growth in customer queries without scaling its contact center at the same pace.
What they built
AI.g, a generative AI virtual agent launched in May 2023. It takes natural-language questions about topics such as bookings, flight status, baggage, check-in, frequent flyer awards and lounge access, and escalates to contact center staff when needed.
How they did it
The core language engine uses Azure OpenAI Service, with retrieval over Air India content through Azure AI Search and Azure AI Content Safety screening harmful content. Conversations, user state, vectors and booking data sit in Azure Cosmos DB, and the bot code and NLP service run on Azure App Service. Customers reach it through the website or mobile app, and it integrates with back-end systems such as the reservation system.
Results
Microsoft reports nearly 4 million customer queries handled, 97% of them with full automation, and about 30,000 questions a day across about 1,300 topics. It also reports savings of several million dollars per year. In June 2024 afaqs reported over two million queries and a 93% containment ratio, meaning about 7% of queries went to a human agent.
All reported figures
| Measure | Figure | Citation |
|---|---|---|
| Customer queries handled with full automation | 97%1 | |
| Customer queries handled | Nearly 4 million1 | |
| Questions per day on average | About 30,0001 | |
| Topic areas covered | About 1,3001 | |
| Containment ratio reported by Air India in June 2024 | 93%2 |
Lessons for your project
Use retrieval over policy documents so the assistant answers from approved sources, and keep a clear escalation path to contact center staff.
Add a content safety layer in front of a public-facing generative assistant.
Connect the assistant to back-end systems such as reservations so it can do more than answer FAQs.
Caveats
- Automation figures are reported by Microsoft and Air India; no independent audit was found.
- The 97% (November 2024) and 93% (June 2024) automation rates come from different dates and may be counted differently.
- Microsoft says the model was updated to the latest GPT models over time, so no single model version applies.
Sources
- Air India elevates customer support while saving money with Azure AI, data, and apps
Microsoft · 2024-11-15 · Vendor case study · microsoft.com
- IndiGo and Air India enhance customer care services with AI-based tools
afaqs! · 2024-06-26 · News · afaqs.com
Summarized in our own words from the sources above. Spotted an error? See corrections.