Build an in house airline support agent that customers can choose over a person

Air India's in house team built AI.g, a customer service agent on Azure OpenAI, after millions of queries pushed up costs and wait times. It went from first work to answering customers within six months, and the team kept improving it by studying real conversations. It now covers more than 1,300 question types, from booking changes to refunds, and handles about 40,000 queries a day. The airline says it has resolved over 13 million conversations with 97% success, so about 3% go to a human. Customers can still pick a person, and about half choose the agent first. Next, the team is testing agents that run whole refund workflows across systems.