In mid-2016, Manish Singhal and Umakant Soni, both entrepreneurs in previous avatars, set up pi Ventures to invest in early-stage startups that use Artificial Intelligence (AI), machine learning (ML) and the Internet of Things (IoT) to solve problems across a range of sectors. The $30-million (around 190 crore) fund, which announced its first close at $13 million in March this year and received another $3 million infusion from the World Bank Group's International Finance Corporation in May, will look to invest in 18-20 startups in India over the next 3-4 years, according to Singhal. He speaks to Forbes India about his game plan. Edited excerpts:
Q. What led you to set up a fund with a focus on AI, ML and IoT?
Our thinking at the start was simple. Data is becoming an integral part of every product. We realised that products of the future will be the ones that are able to utilise intelligence from data very well. Take the case of Facebook. When you use it, it picks up photographs, identifies people and gives you relevant content. Or when you use Google Maps, it is figuring out traffic patterns, where you should go, etc. So products which are using data intelligently will become market leaders.
And the best way to use intelligence in real time is through AI and ML. In conventional, rule-based analytics, there are 'if' and 'else' kind of algorithms-if this happens, then what? But the grey areas that lie between a clear 'no' and a clear 'yes' are hard to make out. With AI and ML, which try to mimic human perception, the learnings are pattern based, so they do a much better job of plugging those gaps. Basically our initial insight was that data is going to be king, or the new oil as they say. And so, AI and ML form the right hand pump, if you will, to take out that oil.
Our second insight was that it's a highly specialised field. You need to really understand what aspect of AI and ML to use and how to use it. We believe that with our background in technology we can work with startups more effectively to bring that out. [Singhal co-founded LetsVenture.com, a marketplace for startups and early stage investors. He draws on rich product experience having worked with Sling Media, Ittiam Systems and Motorola, and has written a part of the MPEG-4 standards, the dominant video standard even today. Soni was director-India, Science Inc, and co-founder of Vimagino, one of the early AI bot companies out of India. Both are IIT Kanpur alumni.]
Third, we are moving from a mobile-first to an AI-first world. Ten years ago, we were asking startups 'are you on the cloud?' Five years ago the Question was, 'are you on mobile?' Just give it two years, and you won't even have to ask whether you are using AI or ML because products that don't use these technologies will die a natural death.
Q. AI has become a bit of an abused term today. How do you distinguish between startups that are using AI and those that are pure data analytics-based companies?
That is one of our core skills, we are technologist entrepreneurs. We deep dive into the technology aspect of the startups to distinguish whether they are actually using AI or not. We look at their product architecture, their algorithms, data strategy, everything.
Even within AI, we look at several aspects: How defensible is it? [That is, does it have competitive moat?] How repeatable is it? [Will it give the right results every time?] How transformational is that AI? So then, there is a lot more gradation there, rather than just being no AI or AI.
Q. What criteria do you use to evaluate investments?
The normal criteria of venture investing still applies-good team, good business model, market size. In addition, we consider two more parameters.
First, IP [intellectual property]-led products are a key criteria for us along with the business. In terms of algorithms, we look at whether there is defensibility, and whether there is applicability. For instance, this algorithm by SigTuple [one of pi Ventures' investee companies that has developed a tool for the automated analyses of blood samples], how good is it to do the blood pathology work they are doing?
Second, we look at data strategy deeply. Since AI feeds on data, unless the startups have a good data strategy, they will struggle. So why did it take SigTuple over a year to develop its IP? Because they needed data. When they started out, they realised that there is no digitised blood data available. That's why they came up with a contraption on a microscope to first produce data. So data strategy for us is critical. We ask Questions like, how do you get access to data? Does your solution naturally produce the data which you can use? What is the value of that data?
Q. Tell us about the investments you have made so far.
We've made four investments so far. SigTuple, which has developed a tool to diagnose blood through image analysis and AI, was our first investment. Their solution takes away the need of the pathologist to physically look at a slide through the microscope and also reduces the cost of diagnosis to a fraction of the current cost. They also do semen analysis, urine analysis and retinopathy. We took part in their $5.8 million Series A round led by Accel Partners.
ten3T is another company where we did a seed round of about $250,000. They've developed a smart patch, which a patient wears on his chest and from which data is transmitted in real time to the cloud. So, a cardiologist can get a full medical grade ECG on her tablet wherever she is. This allows for continuous monitoring of the patient.
We led a 8 crore investment in Zenatix along with Blume Ventures. The company provides an energy management product that helps building managers save up to 30 percent on their energy cost using intelligence from real-time data.
Our fourth investment [a seed round for an undisclosed amount] is in Niramai, a startup that is in the breast cancer screening space. They have built computer vision technology, which works on thermal images to detect early signs of the cancer.
We try to go in as early as possible, typically writing about $500,000 [around 3 crore] in the first cheQue. But it could go as low as $100,000 or as high as $1 million in the first cheQue. We also do follow-ons till Series A.
Q. Three out of four of your investments are in the health care space. Why the focus on health care? Are AI-based solutions more conducive to the sector?
Yes, AI is conducive to health care. Health care has data and large problems to solve, especially in a country like India. We have a population of over a billion people, but very few Quality health care resources. If you fall ill, you typically ask your family or friends about a good doctor to consult. You don't have the confidence to just walk into a hospital and know that you will get Quality care; forget about getting Quality care in tier 2 or tier 3 cities. Think of it as a pyramid where on the top there are very few Quality resources and at the bottom there is a massive population base that needs access to Quality resources. Clearly, there's a wide demand-supply gap. That gap or the middle layer of the pyramid can be filled with AI-like solutions. With Niramai's solution for instance, a technician can do a pretty good job at breast cancer screening; you don't need an oncologist with 15 years of experience. Quality health care can scale with technology, and AI will help with that.
Q. While AI is an enabling technology, what hurdles do you foresee in its development in India?
In India, I would say three things are coming together that are triggering the adoption of AI. They are actually three strong points, but wherever there is a strong point, there is a challenge. First is the availability of data. Second is availability of talent. But that is changing. American Express, Amazon and Microsoft had set up their data science centres in India a few years ago; that talent is now coming out and either joining startups or becoming entrepreneurs. So data scientists are a lot more available now. Third is the adoption of AI by businesses. Are Indian businesses willing to adopt AI to leapfrog? We have actually seen good traction on all three points. So I would say more than challenges, [the ecosystem] is developing right now. It'll take probably 2-5 years, to get into a steady state where we will see check marks on all these pointers