Bengaluru, July 30 -- A service pitched as a way to build trust instead had sparked a debate over privacy. In May, when Bengaluru-based home services startup Pronto piloted an optional service in which workers wear cameras while cleaning customers' homes, critics questioned whether recording inside private homes, even if with consent, went too far.
Nonetheless, Pronto has doubled down on scaling Verified, the opt-in service that enables physical AI data collection, which it says has "anonymized" footage that trains artificial intelligence (AI) systems for robotics. The quick home services platform now bets that the service will account for a chunk of its revenue over the next five years, founder Anjali Sardana said.
"The physical AI data business and our home services business are not two different verticals. It's one. They're highly synergistic in that our services provide the raw material for the data business," said Sardana in an interview to Mint.
To be sure, the home services startup isn't the first one to make physical AI data collection part of its business, it's only the first in its sector. Unlike most physical AI collection services, Pronto does not plan to sell the data to a third party to clean up; it will go in for in-house data collection, annotate and sell directly.
"Our margin on Verified currently swings between 20-70%, primarily because of cloud computing costs. We're expecting that to stabilize in a month or two to around 50-60%," said Sardana.
The founder said that due to Pronto's aim to keep its data free of personally identifiable information and privacy concerns, all data post processing is largely done on graphic processing units (GPUs), without human intervention, which is driving costs up.
That is also the reason the firm is hiring machine learning and reinforcement learning engineers and data scientists to help make data set processing more efficient and cost-effective.
Over the next five years, Pronto expects its data business to account for roughly 90% of its revenue, at a gross margin of around 50%. The firm was set up last March and is yet to file its earnings.
Verified puts Pronto into the sparse, but competitive, physical AI data collection race. With AI in robotics being viewed as the next big tech frontier, labs find there are not enough datasets to train physical intelligence models. This has triggered a race to collect data or provide services related to real world data collection.
"Companies are trying to get more and more data about the physical side of human life, so that they can create solutions that can impact them, and that market is multiple times of the digital market," said Amit Khanna, partner and automation ecosystem leader at Grant Thornton Bharat.
In this context, India's large population of gig workers provides an untapped source of first-person centric data critical to building world models. Globally, there are only a few labs that work on physical AI, including US-based firms Scale AI, Surge AI, Mercor, Turing and Lachy Groom's Physical Intelligence.
Groom, in fact, is an investor in Pronto and wrote a $20 million cheque as part of its $45 million Series B that valued the firm at $200 million.
But not everyone is convinced that bringing physical AI collection to the home is as easily scalable as in a factory, where the variables are limited and more in control.
"In home services, these variables cannot be managed and controlled as effectively as in an industrial setting. Scaling responsibly will not be easy. Companies must look beyond the business model and redesign their operating model to protect the privacy rights of individual customers whose home environments are being captured," said Bharat Chadha, partner at Uniqus Consultech.
Pronto's rivals Urban Company and Snabbit have so far said they will not be recording in consumers' homes. On the other hand, in the US, Shift, a startup based out of New York, offers free cleaning if the workers are allowed to record their work.
While Pronto initially paused the rollout of Verified after the outcry, demand for the service has quietly grown. After a successful pilot, it launched Verified last week, saying that its initial thrust was to bring a layer of trust to customers through the recordings.
The numbers look promising. "Two weeks ago, we were doing 15 orders of Verified a day, today we do about 700," said Sardana. "In early August, we'll be rolling out more devices for our partners to scale Verified." Pronto currently executes a total of about 50,000 orders a day.
The company now operates 130 head-mounted devices with its partners across Bengaluru and the National Capital Region. By August, the number is expected to grow to 1,300 devices.
Verified is currently priced the same as any other hourly booking for a gamut of home services ranging from standard sweeping and mopping of the home to kitchen prep and even helping pack/unpack suitcases. Most services, including those under Verified, are priced Rs.59 for a 30-minute booking, while an hourly slot is for Rs.99. Cleaning of kitchen cabinets, wardrobes, fridge, etc. are priced higher.
For the opt-in service, users need to consent to the recording before paying up. The consent screen says: "Anonmyized videos may be used to help us improve our service and train AI." If they wish to opt out after paying for Verified, Pronto reschedules the service to non-Verified.
And Verified is optional for workers too. Eligible workers, those with enough experience and a qualifying score, undergo three days' training, including on camera use.
The payout is significantly higher for them. "A 1-hour Verified booking pays 2x the standard, a 2-hour booking up to 3x," Sardana said. Published by HT Digital Content Services with permission from MINT. For any query with respect to this article or any other content requirement, please contact Editor at [email protected]