How SpiceJet automated customer support at scale with Sarvam

Background

SpiceJet is one of India's leading low-cost airlines, connecting cities and communities across the country. Its commitment to making air travel affordable and accessible also shapes how it supports its customers.

Customer support is one of the most demanding parts of running an airline. Passengers call with questions about flight status, bookings, refunds, cancellations, baggage, and airline policies, and they expect immediate responses.

To handle these enquiries at scale, SpiceJet built a voice agent powered by Sarvam's Speech to Text and Text to Speech models.

The Problem

Before the voice agent, reservation agents handled customer enquiries manually. Even routine requests required agents to retrieve information from multiple systems and databases before responding.

This became difficult to sustain as call volumes grew. Many enquiries were repetitive and came in Hindi and English, across a wide range of regional accents reflecting SpiceJet's diverse customer base.

Wait times increased during peak travel periods and IROP ( events, such as delays, preponements, and cancellations, when call volumes spiked. Agents spent significant time on requests that did not require human judgement, while expanding the support desk meant adding more people and cost.

SpiceJet needed automated support that could operate 24×7, provide fast and accurate responses, reduce the burden on agents, and scale without a proportional increase in manpower.

Why Sarvam

For a voice agent, the speech layer is critical.

After evaluating multiple vendors, SpiceJet chose Sarvam based on two key factors:

1. Accuracy: Reliable recognition of Hindi and English across a wide range of Indian accents, along with clear pronunciation of airline-specific terms such as "PNR."
2. Cost: Speech-processing costs that remain viable across the hundreds of thousands of call minutes SpiceJet handles every month.

Together, these made Sarvam the right foundation for automated voice support at airline scale.

The Solution

SpiceJet integrated Sarvam Saaras as its Speech to Text model and Sarvam Bulbul as its Text to Speech model within its existing voice agent infrastructure.

When a customer calls, Saaras accurately transcribes the conversation, a language model interprets the request and retrieves the required information, and Bulbul delivers a natural, spoken response back to the caller.

Available 24×7, the bot handles routine enquiries including, booking details, flight status, refunds, baggage information, and (irregular operations) IROP scenarios such as delays, preponements, and cancellations.

By resolving these requests instantly, the bot reduces the burden on reservation agents and allows them to focus on more complex customer needs.

The Speech to Text model is expected to process more than 269,000 minutes of customer conversations, while the Text to Speech model is expected to convert over 269 million characters of text into speech in real time.

Looking Ahead

Over the next 12 months, SpiceJet expects the voice agent to handle a growing share of its support traffic.

As adoption increases, the airline expects to reduce average handling time and staffing dependency during peak periods, while scaling customer support without a corresponding increase in headcount.

The goal is to improve customer satisfaction and operational efficiency while providing fast, consistent support around the clock.