Jul 11, 20235 min read

InsurTech Ohio Spotlight with Jacob Grob

Jacob Grob is the Chief Revenue Officer (CRO) at Tensorflight, a computer vision company extracting data structures from images. Jacob was interviewed by Matt Workman, Senior Client Partner at Persistent Systems.

Jacob, how has data sales changed throughout the years in the insurance industry?

“There's been an influx of data beyond anything that anybody could have dreamed of in the past. There are many organizations that are bringing new data points to the space that have never been available before. Data sales on building characteristics or data about structures themselves have historically come from tax records in the US and ordinance surveys in the UK. That data is lacking in a lot of ways, mostly in direct data population and geographic coverage.

You have some states that are fantastic in coverage and others with poor attribution in their tax records, which makes the hit rate go down. This is something that carriers have been using for years. It's the baseline used by property carriers, firms like Tensorflight, and other computer vision companies are coming in, filling that gap and saying, ‘How do we fill that gap, improve your streamlined processing and allow you to touch less policies? How do we help achieve the questionless quote?’”

What’s the gap in insurance data, and how does AI help augment this data set?

“It’s the gap between the need for data and its availability. There's a core set of elements that every property carrier needs in order to quote a business. When you're trying to streamline quoting and policy admin systems, there’s a finite coverage that's available. The gap comes where that data is not available. What the new AI revolution is bringing to the table is saying, ‘Any building we can see, we can return the building information.’

We can change the paradigm for a carrier where they were only able to quote 40% of policies without touching them. In Texas, computer vision allows carriers to do more properties. It allows for more efficient quoting, a better customer experience and closed business because you are able to provide that quote instantly without having to go back or engage with the agent or insured.”

How does enriched data lead to more accurate quotes?

“If you don't have data and a trusted source, you're assuming the homeowner or business owner knows that data about their structure or building. That isn't always the case. You need a third party source that can validate the information and provide it back to the carrier. This also goes for replacement costs. Undervaluation and ensuring the information that’s associated with policies is vastly different from the structure in real life. Improving the accuracy and the data is what's going to allow us to more accurately quote.”

What are some of the barriers to adoption?

“There's other parts of the data spectrum that don't apply to this, but in the building characteristics, that data from tax records is commoditized. It’s incredibly inexpensive, and carriers are used to that price point. Then you have the gap of 30, 40, 50% where there’s no pre-filled data that’s as cheap as the tax record. It isn't drastically more, but it does cost more.

The value proposition is still within reason. You're not paying dollars per location, but you're not paying pennies either. It's making that concerted effort to say, ‘I'm going to be an organization that provides a more streamlined, easy experience for my insured, underwriters and agents to quote business. I'm going to be more effective because I can instantly provide quotes.’ Making that value proposition to be a leader in this space with that data is where we're at today.”

Where will AI have the biggest impact in insurance?

“It's going to be in the front-end quoting world as we move toward questionless quotes. Ever since I started in the industry, there's always been this drive toward the questionless quote. That's what every carrier wants to get to. They want to be able to put in an address and have all of the data that they need to quote that policy.

We at Tensorflight and those in the data realm are serving this market and given these little pieces of information that allow us to get to a questionless quote. We’re creating a population of data where we have information on every single structure, business and building to be able to answer those questions and fill all of the boxes. It's going to get to a place where you call up your agent and give your name and address with no other questions that need to be asked.”

How are you looking at aerial resolution to overcome the trust factor in the market? What are your thoughts around different methodologies for collecting interior data points, and how do you view AI being a part of that?

“There's always the trust of the imagery and data that's being extracted from the imagery. We are talking about AI models that are not infallible, so we will never get to a hundred percent, flawless coverage. As underwriters start looking at the data next to the image and agreeing with it, a lot of firms are going to as well. It's going to take a little time to build that trust and understanding. When I was first starting out, wildfire risk score was a new thing, and everybody was unsure if it would work.

It took a lot of work with the data, seeing the output and post event to get there. The market is still digesting this information. In some cases, it's done great things to promote AI, but on the same token, it has created a situation where there’s a trust factor that we haven't gotten over. You can look at a property that has been hit by a hailstorm that’s absolutely trashed, but from the aerial image, it looks beautiful. So, it's going to get a great roof condition score even though it has missed the major factor of the hail event. AI has pushed us forward in understanding there's these new data elements that we can start rating, but it also has this blind spot that hasn't been cleared up yet.

Structural information is more clear. Underwriters and risk managers understand that more. That's a 2000 square foot house, right? They can look at that, validate it and move forward with trust in that data. That's going to take time with new data sets. It's going to take energy, but we're going to get there. We're carrying the torch forward on that front.

Now, when you talk about the interior, there's a whole other level of AI. I talk to people all day about AI and what our capabilities are. Today, we're limited to the outside of the building. We need to get inside of houses to start taking it to the next level. There's privacy concerns we need to address, and it's going to require us being invited into the insured's property. As an industry, we need to build that trust with our customer base.

There's definitely a push there, and we're talking to some companies that are looking at technology and interior of the house. Multiple Listing Service (MLS) could also be a place where you can extract some of that data, but the problem with MLS is you get some professional photographers whose job it is to make that house look amazing for sales. Then it doesn’t look like what I saw.

People talk about MLS as being this great source of data that we can train to extract a whole bunch of data. It's not a silver bullet that we think these pros like AI are hard. It is hard. There are many outlying situations that you have to train, understand and compensate for. It's not as easy as, ‘There's a building. I can measure the outsides.’ There’s complexity behind it, and it will take time.”