Air India resolved more than 13 million customer conversations at a 97 percent success rate with AI.g, a generative AI virtual assistant built on Azure OpenAI. The assistant now handles about 40,000 queries every day, has saved the airline millions of dollars, and delivered its first results within six months of development start. The figures come from an official Microsoft post published on April 28, 2026, and from Microsoft's dedicated customer story on the project.
The situation before: a legacy carrier under the weight of its own queries
In January 2022, the Tata Group returned Air India to private ownership after 69 years under the state, launching one of the most ambitious turnarounds in commercial aviation. The airline inherited a brand with deep heritage and a support operation under severe strain. According to Microsoft's case study, the carrier faced “millions of customer queries, which were overwhelming our support channels,” in the words of Dr. Satya Ramaswamy, Air India's Chief Digital and Technology Officer. Rising costs, slowing response times and mounting frustration for customers and staff defined the daily reality of the contact centers.
The classic playbook for a situation like this prescribes more agents, more outsourcing, more call centers — a linear cost curve chasing an exponential query curve. Air India's leadership read the moment differently: a turnaround creates a rare mandate to skip an entire generation of technology, and customer support became the proving ground.
The decision that made it possible: generative AI as the front line, built in-house
Air India became the first airline worldwide to deploy generative AI for customer service at scale, a distinction Microsoft highlights in its April 2026 post. Azure OpenAI reached the market in November 2022; within six months, Air India's internal development teams had built an agentic AI solution on top of it and moved it into production. The company's own newsroom announced AI.g as the airline industry's first generative AI virtual agent.
Three implementation choices stand out. First, scope: AI.g covers more than 1,300 distinct question types, from booking changes to refund requests — high-frequency, well-bounded tasks with documented answers. Second, ownership: internal teams built the assistant on a hyperscaler's model platform, keeping domain knowledge in-house and iteration cycles short. Third, positioning: Air India treated AI.g as the primary channel for customer contact instead of a side experiment. Today about half of Air India's customers choose AI.g as their first point of contact with the company.
The result — 13 million conversations, read with full context
Since launch, AI.g has resolved more than 13 million conversations with a 97 percent success rate. A mere 3 percent of queries escalate to a human agent. Daily volume stands at about 40,000 queries. Microsoft's post states the assistant “has saved the company millions of dollars”; leadership keeps the exact figure private.
SAGA reads success metrics with care, and this case deserves full context. The figures originate from Microsoft, the technology vendor, and from Air India itself — self-reported numbers published through official corporate channels. Independent audits of the 97 percent figure have yet to appear in the public record. The success rate also carries a precise, narrow definition: a conversation counts as successful when it closes with zero human escalation, which measures containment more than customer satisfaction. Read this way, the verified core remains impressive: a documented deployment, a multi-year track record, 40,000 queries per day, and a build time of six months from start to production. Dr. Ramaswamy's team also freed employees “to focus on contributing at a higher level”, shifting human agents toward the complex cases where judgment creates the most value.
What other organizations can learn
The transferable lesson starts with urgency as an asset. Air India's turnaround created an executive mandate for a generational leap: a carrier with 90 years of history rebuilt its front line of customer contact in six months. Legacy status describes a starting point; transformation speed remains a choice.
The second lesson concerns where to aim. Customer queries at Air India's scale — tens of thousands daily, spread across 1,300 repeatable question types — form ideal terrain for generative AI: high frequency, bounded scope, measurable outcomes. Organizations earn the fastest returns by pointing AI at the largest, most repetitive stream of demand they own.
The third lesson is architectural: design the escalation path first. The 97 percent success rate works because the remaining 3 percent lands with a human agent by design. Containment and empathy operate as one system, and the handoff is part of the product. Replicating the result requires specific conditions: query volumes large enough to justify the investment, a clean and current knowledge base for the assistant to draw on, executive sponsorship strong enough to make AI the primary channel, and the discipline to measure and publish success rates. Air India shows the destination: a customer operation where AI resolves the routine at scale and human expertise concentrates on the exceptional.
Article by SAGA — Success Stories & Real Cases
SAGA covers enterprise AI implementations with verified outcomes. Every metric is sourced. Every company is named.