Q&A: Tensorflight’s Co-Founder, Zbigniew Wojna, on Transitioning from Academia to Startup
“Hemi provides us with a very different mindset than ours, coming from research. Hemi helps us to be more forward thinking when it comes to scaling our business.”
Tensorflight, a Hemi portfolio, provides insurance companies with actionable insights from aerial and street view imagery. The company uses building records, geospatial imagery, and machine learning algorithms to help underwriters with risk assessment and to better understand their portfolio exposures.
Hemi first led the seed funding round for Tensorflight in the fall of 2017, with Hemi’s Amy Gu joining Tensorflight’s board of directors. Most recently, Hemi participated in another round of funding as Tensorflight raised $2 million that will be used to scale its computer vision risk assessment service for commercial property insurers. Tensorflight co-founder, Zbigniew Wojna, spoke with Hemi Ventures about his experiences on transitioning from research to entrepreneurship.
What is your personal mission that led you to found Tensorflight?
We really wanted to build something big and useful. Coming from research, I’ve seen many projects that are extremely innovative, novel, and interesting, but to make it really useful, it requires time and contributions from clients, investors, researchers, PMs, and salespersons. That’s why Tensorflight, as a real commercial product, is so interesting from many angles.
At the same time, this problem is very challenging from a technical standpoint. Our software developers are never bored because there’s a new challenge every day. We need to solve many problems simultaneously: scalability, latency, ML issues, computer vision, comfortable frontend, ability to handle large dataset, etc. This is what really drives us.
Lastly, it’s an industry that aligns with our own interests as we have had to insure places to live and to work. Everyone needs a place to live, and every property should be insured. Homes and offices need to have protection from major accidents, worker damage, and natural catastrophes. If the insurance industry is well-optimized, insurance will be cheaper for the end clients, including us.
What makes you excited to go to work every day?
Our team is what excites me. We are all friends who like to spend time together, and everyone is extremely motivated. They make it easy to want to go to work the moment I open my eyes in the morning. There’s a lot more of course — from clients to research ideas, new technologies, and product development. It’s everything combined.
What would people be surprised to learn about you?
Travel is an important part of my life. I aim to visit seven countries each year so that I can cover all the countries before I turn 70. I’ve traveled to 50 countries so far, which is about a quarter of the countries in the world.
Tensorflight is the first company to provide property data on-demand and the first to integrate different imagery sources. How did you achieve this?
Two years ago, we entered the industry, saw what was missing, and knew we could build it. We heard from our insurance clients what they needed and what could add value. The data providers in insurance take the entire state and process the data internally. They respond to direct requests from the database as they come. As a result, the response time will be super quick for most of them (milliseconds), but there are drawbacks. It doesn’t include any new information like new imagery, data sources, etc., and this is the most crucial information for most cases. New houses, new buildings, and all other changes are captured by Tensorflight. This information isn’t captured with traditional methods.
The second aspect is that they usually require purchasing data from larger companies. For example, if you want to process Florida state, you must buy data for the entire state of Florida, which is expensive and challenging. Tensorflight’s technology has the ability to connect on-demand to different data providers (geocoders, symbolic data sources) in one logistical system to shed new light on this problem.
We believe there is a massive market opportunity in computer vision, especially in the insurance space. What do you think are the biggest opportunities?
The major opportunities for this sector are in underwriting process and analytics. The goal for a large-scale client is to provide them with vast amounts of data that they can use to optimize the process. The biggest challenge is accuracy. In computer vision, scaling makes a big difference. In analytics, the underwriting process is more scalable than claims because almost every building is insured and must be renewed every year. Automation can make it better and more reliable.
AI and computer vision are significantly improving the previous manual in-person property inspection process. How do you think this industry will evolve in the next 3 to 5 years, and what are other much-needed automations in this space?
It is not a zero to one transition. Computer vision is a long-lasting field in computer science that took years to develop. In the past few years object detection more than tripled in terms of accuracy. This technology can be applied across the board on a deeper level like in self-driving cars, analytics, and medical imaging. At Tensorflight, our goal is to solve the property inspection problem.
One of our side projects was a hackathon that we won. The Department of Pilots in the Polish government wanted a system to x-ray trucks on borders to look for smuggled cigarettes. Our system’s accuracy level was way beyond other submissions. We prepared systems that recognized cigarettes using pixel-level artifacts.
You have done computer vision research, Google Deepmind, and Facebook AI Research (FAIR). How was your experience transitioning from research to application, from academia to startup?
The non-scientific aspects of running a company are the hardest. The biggest challenge for us now is the opposite of research. We need to be human-oriented so people can trust us and we can prove that our systems are better. It’s the other end of a spectrum that we, as founders, need to learn. We’re not total nerds, but I’m not as good on the business side as I am on the software side.
Hemi led the seed round in 2017. Now the company has grown significantly and is working with leading property and insurance companies around the globe. What contributes to such fast growth?
We want to grow quickly, but working with large corporations takes time. The fastest growth comes primarily from the technology. We also exploited our origin in terms of the formation of the team. Being from Poland, we know that there aren’t as many opportunities to work in exciting startups for engineers there, so we are filling this gap. It would cost us much more to hire such highly-skilled engineers in Silicon Valley.
Big companies are launching “underwriting automation” initiatives (MunichRE, SwissRE, Berkshire) and many commercial insurers are launching “aerial imagery” initiatives. How do you plan to integrate your business with that of larger corporations?
Large corporations have different incentives and are forward thinking players, so it helps to work with them to grow the industry together. We have the best VCs and strategic insurance companies helping us. Over time we would like to be viewed as the gold standard to set up new ways of property inspection. That’s what we’re aiming for, but it’s a partnership that must come from both sides — us and the insurance corporations who are willing to share their knowledge, data, and expertise to push the industry forward.
Smaller startups develop much faster. When there’s a potential for change, it often comes from smaller companies who have been acquired by bigger players that can stimulate their culture. We want to grow organically and build tools that wouldn’t exist otherwise.
What’s your favorite memory of working with Hemi?
As our lead investor, Hemi is very focused on the business side because we are the experts on the technology side. Amy is always asking about our sales, and we rely on her advice and tips. Hemi provides us with a very different mindset than ours, coming from research. Hemi helps us to be more forward thinking when it comes to scaling our business.