Multilingual voice AI for Indian call centers: a build checklist
A voice AI line in Indian languages succeeds when it resolves the calls that make up most of the volume, hands the rest to a person with context, and logs every call in the core systems. A build checklist, with results from a line we built that answers in 11 languages.
Intelligent systems2 min read
A multilingual voice AI line works when it is built around the few call types that make up most of the volume, connected to the systems that hold the answers, and designed to hand over to a person cleanly. Language coverage matters, but a line that understands 11 languages and cannot look up a booking still fails the caller.
We built a voice line for a metropolitan transit network that answers in 11 languages. It resolves 70% of calls without a person, handles the top three call types 80% faster, and logs every call for audit in the network's core systems. RBI Governor Sanjay Malhotra also named voice interfaces in Indian languages when he spoke on AI in banking at FIBAC on August 11, 2026. The checklist below is how we approach this work.
70%
of calls resolved without a person on a metropolitan transit network's voice line in 11 languages
Before you build
- Pull a sample of real call recordings and transcripts, and count call types. Build for the top few first.
- List the languages and dialects callers actually use, by region and time of day, from the recordings rather than from a census.
- Map the systems each top call type needs: bookings, accounts, complaints, payments. Confirm each has an API the voice line can call.
- Agree which calls must always reach a person, such as safety reports, distress and legal complaints.
While you build
- Test recognition on recordings from your own callers, with their background noise and phone lines, not on studio audio.
- Let callers switch language mid-call. Many mix two languages in one sentence.
- Confirm anything that changes a record, such as a cancellation or a complaint, by reading it back before acting.
- Hand over to a person with the transcript and what the caller has already said, so they never repeat themselves.
- Write each call, its transcript and its outcome into the core system the call concerns, with a reference the caller can quote.
After launch
- Review a sample of resolved calls every week, by language, to catch wrong answers the caller did not complain about.
- Track hand-over rate and repeat calls per call type. A falling hand-over rate with rising repeat calls means the line is closing calls it did not solve.
- Add call types one at a time, each with its own test set, once the first ones are stable.
The audit log is what lets an operator in a regulated or public service defend the line. When a caller disputes what they were told, the record shows exactly what was said, in which language, and what the system did.
Questions
- How many calls can voice AI resolve without a person?
- It depends on the call mix. On the metropolitan transit network line we built, 70% of calls are resolved without a person, across 11 languages.
- How do you handle callers who mix languages?
- Allow switching mid-call and test recognition on your own callers' recordings, since many speakers mix two languages within a sentence.
- What should a voice AI line log for audit?
- Each call's transcript, the language, the actions taken and the outcome, written into the core system the call concerns, with a reference the caller can quote.
Sources
- RBI Governor Sanjay Malhotra Sets Out AI Playbook for Indian Banks, The Kanal, August 13, 2026
More on intelligent systems
All writingNovember 2026
CERT-In's AI security audit guidelines: what an auditor will check
CERT-In's AI Security Auditing Guidelines, dated November 4, 2026, sort AI systems into four risk tiers and audit them across nine domains. For AI agents, auditors will test whether an agent can exceed its permissions or bypass approval. Here is what to prepare.
October 2026
How do you test an LLM before it goes live in a regulated business?
Test a language model the way an auditor will question it: on your own cases, against written pass criteria, with attacks on its permissions and data, and with a record of every result. RBI's Governor has asked banks to red-team and stress-test AI before deployment and periodically after; five test sets are where that starts.
October 2026
What drives the cost of an enterprise AI project?
The model is rarely the biggest cost in an enterprise AI project. Integration with core systems, data work, controls and the cost of running it after launch usually decide the budget. Here is how to estimate each before you sign.
Tell us which system the AI has to run inside
If that system is not ready for AI yet, tell us that too. An engineer will reply and say plainly whether we are the right firm for it.