Firms are spinning off AI tools, but are clients interested?

House hunting has moved online in India, but the process still needs plenty of phone calls.

Increasingly, those calls are handled by AI. Voice agents can book appointments, narrow down listings, and negotiate with landlords on the tenant’s behalf.

The agents can even close deals on their own, according to Akhil Gupta, chief technology officer of real estate platform NoBroker. He tells Tech in Asia that an AI agent had successfully negotiated a deal with a customer seeking a fully furnished house by counter-offering a semi-furnished option that included help arranging rented furniture.

Tech in Asia was not able to verify the deal.

NoBroker, backed by private equity firm General Atlantic, is one of the biggest marketplaces connecting landlords and tenants in India. It built Convozen, a conversational AI agent platform, by fine-tuning several open-source speech-to-text models on millions of hours of NoBroker’s customer call recordings.

“Fifty thousand hours of telephony-generated data is a lot,” Gupta says. “We took different cohorts, anonymized the data, and used it to train the models. That’s the advantage.”

NoBroker then took the next logical step: commercializing Convozen as an enterprise AI product.

See also: The real money in AI isn’t in the models

But despite launching 18 months ago and onboarding more than 50 clients, Convozen’s annual recurring revenue (ARR) stands at just about US$2 million, according to Gupta, who describes the product’s current contribution to NoBroker’s financials as “small.”

The company’s experiences reflect a growing realization across India and Southeast Asia: Spinning out an AI business is easier said than done. In conversations with over half a dozen executives, investors, and founders, Tech in Asia was repeatedly told that attempting to turn a company’s internal data into a new line of business is very tough, especially in an area where prices are falling, engineering talent is scarce, and competition is high.

Making legacy useful again

When ChatGPT launched in 2022, companies around the world began thinking about how to make their legacy data useful with AI. Most started with customer service – long a high-cost function – by replacing human agents with chatbots and voice AI.

In Singapore, Grab has built AI tools for internal use across its platform, such as an AI driver assistant and a live voice translation feature between drivers and riders.

In India, payments company Paytm is the latest example, having replaced parts of its call center operations with AI agents. It now plans to sell that technology to other businesses.

Data as an opportunity

One conversation for all?

A general lack of interest

Is all this effort worth it?