With Elon Musk and Mark Zuckerberg sparring over its ethics and China announcing its intention to create a $150 billion domestic industry based on it, Artificial Intelligence is perhaps the most discussed topic in the tech news cycle. It’s likely to be a talking point no matter what your favourite watering hole for tech news.

Billions of dollars have been invested by VCs in AI since 2016 with the US and China leading the race in record funding in terms of deals and dollars.

In sharp contrast, Indian startups have collectively raised less than $100 million from (2014-2017YTD), according to data from startup analytics firm Tracxn — that’s smaller than Andrew Ng’s recently launched $150 million VC fund. Another way to look at it: Grammarly, a Valley-based spell check tool raised more dollars than all of India’s AI startups put together in the past three and a half years.

According to a recent PwC report, AI is expected to contribute an additional $15.7 trillion to the global GDP by 2030, with most of the economic gains going to China and US, who will account for 70% of the global economic impact. Does India risk becoming a laggard in the AI race, and what are the potential implications, and existential risks of missing out on this wave?

FactorDaily reached out to VCs and some of the top-funded AI startups in India to hear about the challenges faced in building an AI startup out of India, in spotting AI talent, and finding the elusive product-market fit. We have also got some data on company formation, deal flow, and how it compares to leading nations, like the US, China, Israel, UK, and Canada.

The dice is loaded against startups

Artificial intelligence is an especially thorny space to be in as Indian startups lack access to large data sets. Oxymoronic as it sounds in a country of 1.3 billion, the truth is as simple as that. Tech giants such as Amazon, Google and IBM have larger data sets compared to startups, says Parag Dhol, Managing Director at Inventus Capital Partners, over a phone conversation analysing Tracxn’s data on the AI startup landscape in India.

“To train your machine, you need a large data set. Now, IBM walks into Manipal (Hospital), makes a nice presentation, shows off what they’ve done and gets access to Manipal’s Oncology reports for the last five years,” says Dhol. “If it was a company out of Bengaluru, can it do that? The answer is a no,”

Next is the role of patient capital, which in this instance seems to be coming from governments. US and Chinese government funds are heavy investors in AI companies. In-Q-Tel, the Central Intelligence Agency (CIA)’s VC arm, for example, has backed companies such as Palantir and ThreatMetrix, among dozens of other AI startups. The Chinese government is looking to invest billions of dollars into AI, as a part of its AI development plan, with aspirations to become a world leader in the space by 2025.

“Israel, the US, China… they’re probably so far ahead of us that it is not even funny anymore. To that extent, comparing ourselves to any technology in a latest, bleeding edge kind of area is a fool’s paradise. We need to find a way to do that. Some of these big investments have to involve the government at times,” Dhol adds.

Very few VC funds in India will fund business models with a five-to-ten year timeline where data collection and analysis comes first. For AI companies to proliferate, they need to have their AI wrapped around in a business model, where it makes money to sustain its data collection efforts, Dhol says. “Choose a problem that IBM can’t solve, because the data set is not available to IBM or is uniquely Indian, and go after it. Build and get in some business model where it is very hard to get some data. There has to be some revenue, some profit as you go along. It can’t be five years later,” he says is his advice to Indian AI startups seeking funding.

Despite the adverse environment, there are niches to exploit he said, highlighting Tricog Health, one of Inventus Capital’s portfolio companies. Tricog offers cloud-connected ECG devices, which can detect heart complications within minutes. “ECG data is immensely valuable to somebody. But that requires going out into the field and collecting the data and that has a cost associated with it. So, the path that Tricog Health is taking now is to have a business model where while collecting the data, they make some money,” he says.

The Grass is Greener…

One thing sorely missed by AI startups in India is the network effect. Chip manufacturing and semiconductor design didn’t ramp up in India in a big way and AI insiders worry the same fate will hold true for AI in India. “We don’t have the network effects that are required for even big data. There are very few companies that have access to big data, whereas US and China have thousands of companies that have access to big data one way or the other,” says K S Sreeram, founder of Chennai-based Clay Labs. His company is currently working on a product that automates software development. In 2006, his company – Tachyon Techologies released the first machine learning based Indian language input system called Quillpad.

“You cannot get into chip design, or get access to the core YouTube database, or Google search database. All those are proprietary databases. We are structurally at a disadvantage as a country because the deep tech ecosystem is not there,” he says.

“I don’t think India will be a good market for AI even in the future,” says Vinay Kumar, co-founder of Arya.ai, a Mumbai-based AI startup. Arya provides a deployment stack for deep learning platforms to enterprises with clients in the enterprise space — banking, insurance companies, and manufacturing. Enterprises use Arya’s platform for multiple use-cases, from fraud detection to intelligent automation. “I don’t see there are any advantages of being in India. India is not a big market for enterprise AI yet. We are primarily an enterprise-focused business, for us, the bigger markets are Europe, US, and even Singapore and China,” he says.

As operational costs are lower in India, at least in comparison to the US or Europe, using AI to harness efficiency gains here in India are a low-priority for enterprises, Kumar adds.”Take insurance as an example. To process the claims, or process the house insurance policy, the human resource costs are very high, while In India, I could hire a large number of doctors to look at the documents.”

India is a great place for AI startups to experiment, but not a lucrative market, says Ashwini Asokan, co-founder of Mad Street Den, a AI company from Chennai that’s focused on visual recognition. “Supply and demand in India has been non-deep tech. To me this is not surprising at all”, she says about Tracxn’s data on investments in Indian AI companies.

“We started in 2013/14, Nobody knew what computer vision was in 2013 and our application-focused approach to building our tech helped. It allowed us to go to market and experiment very quickly. But if you don’t figure out how to get out of the Indian market really soon, it’s instant death. B2B does not pay in India. In B2B speak, you can’t hunt elephants in the Asian market. You can hunt flies, mice and rabbits at best,” she says.

Mad Street Den had to pay for these early experiments out of its own pocket, on its path to finding a product-market fit. “We wrote down a lot of bad debts from this side of the world,” she says.

To be sure, India’s startup ecosystem has experimented a lot in sectors such as logistics and payments. But not in deep tech sectors, Asokan says. “I’d say both deep tech as well as B2B are not the majority in India. B2C is. That’s on the startup side, and obviously, on the investment side, you see that reflected. Even within B2B, it’s not common to see enterprise startups with greater than $5,000 a month price range.”

With the new wave of deep tech startups that are coming up, the VCs are very cautious, as there are no proven winners emerging out of India, she says. “I would argue that any kind of investment around deep tech in India is a massive experiment, and the rounds are a lot smaller than they would be in the US because we have no history, no template to follow through.”

Big data needs big government

Ironically, India is at the centre of several big, global AI initiatives. “India has the world’s largest number of Android users, they are contributing to Google’s data moats, which in turn drive their AI efforts,” says Vivekanand Pani, co-founder Reverie Language Technologies. The Bengaluru startup provides a language-as-a-service platform, used by Indian companies to support their localisation efforts.

“Availability of data sets that could be used to train AI engines is a pain point that the government could alleviate through laws on data privacy and data accessibility”, he says. This sentiment was echoed by Kumar as well. “India would definitely need to think a lot about data privacy and open data platforms. There is a lot of health data in India, but it is not being offered to Indian startups,” he says.

“I think there’s a lot of effort and energy required for thinking about policy if the Indian government wants to do anything with AI,” says Asokan. “I’m not certain the Indian government wants or finds the need to use AI. I can see security, defense-focused applications. If anything, I’d argue, thinking carefully about the service economy is more important because that’s what India is,” she says.

We discussed the Indian government’s mindset with respect to AI, in light of a recent statement by the Union Transport Minister Nitin Gadkari where he said that India will not allow driverless cars to ply on its roads. “If anything, I’d argue that there will be resistance because there’s a lot of service money, or a service economy that the country is benefitting from. If there is interest, it will have to be an effort very specific to the Indian context, that is lacking in public infrastructure, which is fairly fundamental to using AI systems. So we’ll have to think about this very differently than throwing self-driving cars or automated machines all over cities,” says Asokan.

A case for optimism

“Yes, from a numbers perspective, we are behind. But what to me is positive is that we’re investing in research, in education and training, and retraining, in solutions using AI, and machine learning,” says Kris Gopalakrishnan, chairman at Axilor Ventures, and co-founder of Infosys in a phone conversation where we discussed how AI will shape the startup and IT services landscape.

Companies like Infosys, TCS, Wipro, are training their people on some of these new technologies, he said, and there’s a growing base of AI research that’s happening at places like IIT Madras, and IISc. “If you look at one factor which is normally not tracked — talk to any of the technology companies who have their R&D centres in India, you will find that there’s a growing number of projects on AI which are happening in those research facilities, be it SAP, Oracle, Microsoft…,” Gopalakrishnan says.

“In terms of talent, one thing I could see is that talent is very cheap in India. There are a handful of institutes (the IITs and the IISCs) that have the knowledge and talent pool, and we can find a lot of good people in a very quick time, compared to the US or any other place,” says Arya’s Kumar. “The calibre might be different — for example, in the US, I could find an experienced deep learning expert, but in India, I can only find a fresher or a researcher in a different domain.”

India-specific problems are being addressed, too. “To me what is good is that more and more startups are embedding AI in their technology stack and starting to supply AI for solving issues sometimes typical for India,” says Gopalakrishnan. He gives the example of Sigtuple, a company Axilor funded that has leveraged AI and machine learning to remove the need for pathologists to analyse blood samples — even in small villages.

Also, some data is at hand. Owing to India’s large population, banking or insurance companies have large data sets to analyse. “Many large banks have more than five million accounts. That might be another advantage when compared to other countries,” he says.

Clay Labs’s Sreeram says that startups working on a class of AI that works with small data have a really good future in India, as it can be done without needing access to non-existent infrastructure and network effects. “In this area , algorithms that can learn from a small set of examples — productizing that is a unique opportunity for India because it just comes down to having a breakthrough idea and putting together a small team to do that. There are tons of opportunities in that domain,” he says. He details a use case in customer support where a start is made with automatic replies to a small percentage of email queries which scales up with time. “Even if an AI is able to answer 20% of your incoming queries, it’s an immediate win. You don’t need to hire new people next month,” he says.

India is behind countries like China and the US from an academic research perspective, and hence behind translating that knowledge to startups, and products, Gopalakrishnan says. “But given that we have probably the second largest number of IT professionals in the world, and we’re good at retraining these people, we’re going to catch up soon, or at least we’ll be in the top five.”

“Why am I optimistic? I’ve seen this. India imported its first computer in 1955,” he says, linking me to one of his passion projects (itihaasa.com), which chronicles the history of computing in India. “We were almost on par with the world in realising the power of computing. We created the first master’s program in the 1960s at IIT Kanpur. We trained a group of people who started the first set of IT companies in India in the late 70s and early 80s. We have now four million plus professionals in IT services, we created a $160 billion industry, exported $130 billion, it’s working with all the multinational corporations in the world. Why did it all happen? Because there was a concerted effort to create an industry. It took 62 years. But we did it.”

---