Karnataka IT Minister Priyank Kharge hands an MoU to ElevenLabs CEO Mati Staniszewski on stage at the ElevenLabs Summit in Bengaluru — Karnataka Minister Priyank Kharge hands the MoU to ElevenLabs CEO Mati Staniszewski at the ElevenLabs Summit in Bengaluru on October 6, 2026. Photo: Reuters, via The Hindu.

For years, voice AI in India felt like a demo that never quite reached production. Accents, code-switching and patchy phone lines broke most bots within a minute. This week, one of the biggest names in the field said that is changing, and it is putting serious money behind the claim.

On October 6, 2026, ElevenLabs held its first India Summit in Bengaluru. Its cofounder and CEO, Mati Staniszewski, told Reuters the company plans to invest hundreds of millions of dollars in India. I went through the coverage from Reuters, CNBC-TV18, The Hindu and the company's own summit announcement. Here is what it means for agencies and software teams in India and the USA.

What ElevenLabs Announced in Bengaluru

The headline is the investment. Staniszewski told CNBC-TV18 in an exclusive interview that ElevenLabs will spend a double-digit percentage of its recent fundraise in India. The company raised close to $1 billion, so that works out to hundreds of millions of dollars.

Reuters and CNBC-TV18 reported that the money will go toward three things:

Local teams across go-to-market, engineering, research and forward-deployed engineering.
Model development built with India in mind from the start.
Indian language work, with a plan to cover all 22 officially recognised languages.

Reuters also reported that ElevenLabs is open to mergers and acquisitions in India. The company did not name any targets or share an exact investment figure.

The Numbers Behind the Bet

These figures come from the company, shared at the summit and with CNBC-TV18:

India is now ElevenLabs' second largest market after the US.
Global revenue has crossed a $600 million annual run rate.
Enterprise customers now make up 55% of the business, ahead of self-serve users.
More than 250 businesses in India work with the company.
Indian businesses ran 100 million conversations on its ElevenAgents platform in the past year.
Over 70% of those conversations were in Hindi, Kannada, Tamil and Telugu.
More than 25,000 Indian developers and businesses build with its API each month, up about 2.5 times in a year.

Staniszewski summed up the pitch in one line to CNBC-TV18: "India can be a laboratory for voice AI."

Why Voice AI Is Finally Ready for Indian Customers

Text chatbots work fine for people who type in English. A large share of Indian customers would rather speak, and often in a mix of languages. A delivery partner might switch between Hindi and English in the same sentence. Older bots failed exactly here.

ElevenLabs CEO Mati Staniszewski on stage in Bengaluru in front of a screen reading 22 official languages, thousands of dialects and accents, with sample phrases in Hindi, Marathi, Tamil and Hinglish — Staniszewski presents ElevenLabs' plan for all 22 official Indian languages at the Bengaluru summit. Photo: DQ India.

ElevenLabs says its new Eleven v4 model, launched last week, supports 14 Indian languages today. DQ India reported that the faster v4 Turbo version targets a median latency of about 100 milliseconds for live conversations. That speed matters. A long pause on a phone call is the quickest way to make a caller hang up.

The company also says it has built for local needs that global tools often skip:

Code-switching, so a bot can follow Hinglish or Tanglish without breaking.
Data residency options for clients that need data kept in India.
Telephony links with Indian providers Ozonetel, Exotel and Plivo for inbound and outbound calls, routing and handoffs to human agents.

Real Voice AI Deployments Agencies Can Learn From

Launch numbers are easy to ignore. Deployment numbers are more useful, because they tell you what a realistic first project looks like. These are figures shared by ElevenLabs about its Indian customers, so treat them as vendor claims rather than audited results.

Company	Use case	Reported result
Razorpay	Hinglish outbound calls to merchants	35,000+ calls, about 28% connection rate, on par with human call centres
CARS24	400+ voice agents for inbound calls	3 million minutes automated, 30% of inbound calls handled, costs down 50%
TVS Motor	Voice agents in 25+ countries and nine languages	35% more leads captured, feedback collected 80% faster
Urban Company	Support for gig workers in seven languages	55% of partner queries resolved without a human

Notice the pattern. None of these replace a whole support team. Each one picks a narrow, high-volume task with a clear script. That is where I would start with any client.

Other uses go beyond calls. Pocket FM uses AI voices for long-form audio series. Havells uses speech-to-text and text-to-speech as the voice layer in its smart home app. At the summit, ElevenLabs also showed a single ad campaign adapted into Hindi and Kannada while keeping the same brand direction.

What This Means for Agencies in India

If you build software or run digital projects for Indian clients, expect voice AI requests to rise. A global vendor spending this much locally usually brings sales teams, partner programs and plenty of client curiosity.

Here is where I see real demand forming:

Customer support for D2C and fintech brands that get thousands of repeat calls about orders, refunds and KYC.
Lead qualification calls for real estate, auto, education and insurance clients.
Regional content such as product videos, explainers and ads in multiple Indian languages.
Government and civic projects, which are opening up as states test voice tools.

That last point is already moving. ElevenLabs signed an MoU with the Karnataka Innovation and Technology Society to explore voice AI in Kannada for skilling, citizen services and grievance redressal. ANI reported that the MoU is voluntary and non-binding, and it will not decide any future tenders. So it is a signal, not a contract pipeline.

Karnataka Minister Priyank Kharge and ElevenLabs CEO Mati Staniszewski hold signed MoU documents at the ElevenLabs Summit in Bengaluru — Kharge and Staniszewski with the signed Karnataka MoU on voice AI in Kannada. Photo: Express Computer.

Also keep an eye on public tools. India's government-backed Bhashini language platform works on translation and speech tools for Indian languages. For some public sector or cost-sensitive projects, it is worth comparing against a commercial vendor.

What This Means for Agencies in the USA

US agencies should not treat this as only an India story. A model that handles code-switching, accents and noisy lines at scale tends to get better for everyone. Many US brands also serve South Asian customers or run support centres in India.

There is one legal point every US team must know before building outbound calls. In February 2024, the FCC ruled that AI-generated voices in robocalls count as artificial under the Telephone Consumer Protection Act. In plain terms, AI voice calls fall under the same consent rules as prerecorded robocalls.

So for US clients, I would steer early projects toward inbound support and calls customers have clearly agreed to receive. Get legal sign-off before any cold outbound campaign.

How to Run a Safe Voice AI Pilot for Clients

Whichever vendor you pick, the process matters more than the brand. This is the checklist my team uses when a client asks for a voice agent.

Pick one narrow task. Order status, appointment booking or payment reminders work well. Avoid open-ended advice on day one.
Test real callers, not studio audio. Use recordings with background noise, regional accents and mixed languages.
Measure latency on real phone lines. A fast demo in a browser can feel slow over a mobile network.
Design the human handoff first. Callers must be able to reach a person quickly, with context passed along.
Get consent for voice cloning in writing. Never clone a voice without clear permission from that person.
Check where data lives. In India, map the project against the Digital Personal Data Protection Act. Ask vendors about residency and retention.
Tell callers they are talking to AI. It builds trust and lowers complaint risk.
Log and review calls weekly. Track resolution rate, drop-offs and escalations, then tune the script.

Voice agents are still AI agents, so the same governance rules apply. Our guide on AI agent governance for agencies covers access controls and audit trails in more detail.

Risks and Open Questions

I like where this is going, but a few things deserve caution.

Vendor claims are not benchmarks. The deployment figures come from ElevenLabs and its customers. Run your own tests before quoting results to a client.
Language coverage is still growing. 14 languages is a strong start. Projects in the other eight official languages may need to wait or use another tool.
Pricing can change. Build your cost model per minute of conversation and leave room for vendor price moves.
Voice misuse is real. Deepfake voice scams are a known problem. Clients will want clear rules on who can create and use cloned voices.

Key Takeaways

ElevenLabs plans to invest hundreds of millions of dollars in India, a double-digit share of its recent fundraise of close to $1 billion.
India is now its second largest market, with 100 million agent conversations in the past year.
Eleven v4 supports 14 Indian languages, with plans for all 22 official ones.
Real deployments at Razorpay, CARS24, TVS Motor and Urban Company show voice AI works best on narrow, high-volume tasks.
US teams must respect TCPA consent rules for AI voice calls. Indian teams should plan around the DPDP Act.
Start with one pilot, real caller audio, a clear human handoff and weekly reviews.

If you are weighing a voice agent or any AI feature for your product, browse our AI services and guides or talk to our team about software development consulting. We are happy to help you scope a pilot that fits your budget.

About the Author

Nitin Kumar is Lead Project Coordinator at Debugged Pro. He manages software and AI delivery for clients in India and the USA, from scoping to launch. He writes about practical ways teams can adopt new tools without losing control of quality, cost or data.