From 100,000 calls a month to 5 million a day: Inside Vobiz’s AI telephony bet

When Bengaluru-based Vobiz.ai began operations late last year, companies using its telecom APIs were making about 100,000 AI-generated calls a month by January. Less than a year later, that number has climbed to an average of five million attempted calls a day, as the startup sees customers moving from proof of concept to production.

Founded by Suman Gandham and Vikash Srivastava, Vobiz is betting that the rise of AI voice agents will create demand for a telecom layer built for machine-led conversations, rather than traditional call centres.

"We are now doing five million attempted calls a day and expect that number to reach at least 10 million calls a day by December this year" Gandham, cofounder and chief executive officer of Vobiz.ai, told The Economic Times Digital.


Also read: Enterprise Voice AI in India: How AI-powered voice technology is scaling across businesses

Gandham said that the company has crossed $1 million in annual recurring revenue with around 65% of that coming from contracted business. He added that the company is targeting $2.5 million by December, based partly on a pipeline of companies currently testing the platform, running proof of concept or in commercial discussions.

The bet comes as businesses increasingly experiment with AI-powered voice interactions. According to a report by The Business Research Company, the global voice AI market is expected to reach $32.47 billion by 2030, growing at a compound annual growth rate of 29%.

Gandham added that existing customers are also increasing their usage, with monthly call volumes rising to 237%.

The company's customer portfolio includes companies such as Sarvam AI, Shiprocket, Razorpay, RevRag, Bolna, Navana AI, among others. It competes with domestic and global players such as Exotel, Plivo, Twilio and Telnyx.

Building below the voice agent

Gandham has spent years in the telecom and fintech ecosystem before starting Vobiz. He previously founded click-to-call startup Letmecallu and voice CRM company voz.io in Barcelona, both of which were later acquired. He then moved into venture capital, where he ran a $110-million fund focused on sectors such as fintech, proptech, travel tech and mobility.

Also read: Voice AI goes mainstream as enterprises move beyond pilots

After six years in venture capital, he moved back to India and founded consumer neobank Finin, which was acquired by Open Financial Technologies for $10 million. He joined Open as part of the acquisition.

After leaving Open, Gandham began building AI voice agents for BFSI companies, which eventually led him back to telecom and the idea for Vobiz.

He said the voice agent market became crowded quickly, but while building those products he kept running into another problem.

"The biggest problem I faced while building voice agents was telephony. I wanted to know whether it was a problem only for me, so I spoke to hundreds of people building voice agents. Everybody said they needed good telephony infrastructure," Gandham said.

A typical voice AI stack combines a large language model, speech-to-text, text-to-speech and telephony. Telephony is the part that connects the AI to the actual phone network, allowing it to make and receive calls. Vobiz focuses on this layer.

This led him to Srivastava, who had previously worked at telecom infrastructure companies including Plivo and Bandwidth. The two decided to build Vobiz as an AI-first telecom infrastructure layer.

"Existing telecom companies are legacy incumbents. Their infrastructure was built for human-to-human calls and call centres, not for AI-first communication," Gandham said.

Also read: Business outcomes, not AI models, will decide enterprise deals: Gnani.ai CEO Ganesh Gopalan

Vobiz has built what it describes as a single-hop architecture across its transport, media and socket layers. Gandham said the company records P95 telephony latency of under 80 milliseconds, including the carrier and public switched telephone network leg.

"We want to do for telephony what Stripe and Razorpay did for payments," he said.

Sarvam puts the stack to use

At Sarvam's Epoch event in July, Vobiz showed how it wants its infrastructure to work through its integration with the AI startup. It powered the telephony layer behind Sarvam's Samvaad platform, allowing users to complete KYC, provision an Indian phone number and connect it to a voice AI system.

Gandham said obtaining a number through existing providers can take weeks because of paperwork, onboarding and integration requirements.

Also read: Everything Sarvam AI announced at its Epoch Builder Edition

"Once the KYC is complete, customers can get a DID (Direct Inward Dialing) number through our APIs. We maintain a pool of numbers and a customer can get one in 30 seconds. When Sarvam came to us, its biggest requirement was to make it easy for users to go live immediately," he added.

The integration itself took around 30 to 45 days, Gandham said, as Sarvam tested the platform, account architecture and integrations at scale.

The company also provides integrations with platforms including LiveKit, Pipecat and ElevenLabs.

Gandham said that BFSI is the biggest contributor to Vobiz's voice AI traffic, followed by logistics, adding that e-commerce, healthcare and real estate are also seeing growing adoption.

What starts breaking at scale

As call volumes grew, Vobiz encountered problems that were less visible at a smaller scale.

One challenge was securing enough capacity from telecom operators. Gandham said additional carrier capacity earlier came with a lead time of one to two weeks, which has now reduced to two days. The more unexpected problem was phone numbers being labelled as spam.

According to him, even consent-based calls can be marked as spam, reducing connection rates and prompting customers to ask for replacement numbers.

To tackle this issue, Gandham said, Vobiz has built a Number Health Management system that monitors the performance of individual numbers and allows customers to change them when their health deteriorates.

Also read: Voice AI moves from an add-on to the core for Indian companies

He added that Vobiz does not allow cold calling and asks customers to provide proof of consent when complaints are raised. The company also monitors call patterns for sudden spikes that could indicate bulk cold calling activity.

From a 'dumb pipe' to an intelligent layer

Vobiz now wants to move beyond simply carrying voice traffic by adding more pre-call and post-call intelligence.

"Telephony has traditionally been called a dumb pipe. We want to turn it into the intelligent communication layer for AI, whether the interaction happens through voice, SMS or WhatsApp," Gandham said.

He gave the example of a collections call in which a customer tells an AI agent that the credit card bill will be paid in five minutes. Vobiz wants its system to understand that response and automatically send a payment link over WhatsApp or other channels.

"Our goal over the next couple of years is to become the intelligent communication layer of AI," Gandham said.