The Data Supermarket and it’s Master Chefs

A great interview by Matthew Grant of the excellent Instech London, with WhenFresh co-founder Mark Cunningham. Key talking points include:

> Why insurers should embrace pre-fill
> Getting trusted data during Covid-19
> Identifying new sources and partners
> Moral hazards and customer experience
> The hidden costs in property valuations
> Why “WhenFresh” rather than “ArtificialElephant”

Please listen to the podcast, or read the highlights transcript below.

The Data Supermarket and it’s Master Chefs – Podcast highlights:

Matthew: Mark, you founded your first company in 2006 and sold it in 2011. You also developed a music search library before setting up WhenFresh in 2012. How did you end up in insurance?

Mark: We took the data warehousing and search capability from the music search business and looked at where else we could apply it. My co-founder, Alan Dean, came from an insurance technology background. He was with Swiftcover and Microsoft and Tesco before that. We found nobody else was doing this kind of data rendering in insurance.

The last year has been a sea change in how people use information to make distance decisions. Remember when we used to go and meet people? We can’t do that now, so we need to look for data elsewhere.

Matthew: This concept of WhenFresh being a data supermarket. What are you offering to insurance companies?

Mark: It depends on the company. Typically, a company would say their pricing team needs to know certain things, e.g. proximity to water, a previous flood event, building height. They want to know that the data is right and is it available quickly?

Matthew: Once you’ve got the data, who does the analysis?

Mark: We do it in-house through Alan, our Head of Data Ken Clemmer, and Seb London. Seb was the head of API development at comparethemarket, so we have three expert checkers. Think of it as a restaurant. They’re the chefs creating and checking the food’s right.

Each of the data items comes from a provider, we work with what we think are the best companies, whether it’s satellite imagery out of the University of Liverpool or work from Cranfield University or the European Space Agency.

Matthew: There’s so much free data out there. Are companies providing your paid-for data, and can your team add value from free sources?

Mark: Free data can actually be really expensive. If it’s not properly processed, people make the wrong decision. What we’re looking for is provenance. Can I bind the free data to an address or map reference? That’s tricky and using free data incorrectly can cause real trouble.

Matthew: You’ve done something quite clever with data from Zoopla. What is it that you’re providing?

Mark: They’re a key strategic partner and we extract key value elements from their descriptions of properties. Think of it as two particularly useful outcomes. One where an agent accurately describes something, and one where they accurately describe something that hasn’t been described before. A loft conversion, or a basement or side return that doesn’t appear on planning data or isn’t visible in imagery data. We know someone has sent a human to have a look so that data is really useful in the insurance space.

Matthew: That’s one of the challenges with data, particularly with properties. People make changes all the time which can impact the value.

Mark: Zoopla has been a very good source; as have planning data and imagery data providers. We aggregate those to build a patchwork that gives a much better view. If Zoopla, the planning application and the loan application say it’s a four-bedroom house, then it’s a four-bedroom house.

Matthew: There will sometimes be different views though, so is the challenge making a decision based on the different sources available?

Mark: We have to be careful about how we interpret data. I’m sitting in a room which is designed as an office but was formerly a bedroom. I had my house valued and they told me I had a three-bedroom house, but it’s a four-bedroom house. It doesn’t have to have a bed in it to be a bedroom.

Another example is I can describe a bungalow that’s an end of terrace as also being semi-detached and also a bungalow. Which one’s right? We’ve got to know what the insurer needs, then recode the data in accordance with their flavour.

Matthew: You also provide choice, rather than saying, ‘this is the WhenFresh view, take it or leave it.’

Mark: The insurer drives the decisions and they need to know the source of the data. Going back to the supermarket analogy, a customer has got to be able to walk along the shelves and choose which brand of corn flakes they want.

The essence of our business is delivery in a way that gives insurers confidence. We should be the least exciting company they’ve ever come across, because what they asked for turned up exactly when expected and it does exactly what we said it would do.

Matthew: Are you doing any crowdsourcing of opinions of which data people prefer? Rather like the Netflix model of ‘if you liked that, you’ll like this.’

Mark: The market is small enough that I’m not sure we have the same gravity traction in terms of usage, but it’s a very good point. Two of our biggest customers now give us corrected data where a customer has told them that something isn’t true and we update our model. Using them as an ingredient source is in effect crowdsourcing. A customer with 1.5 million policies helps infer what the truth is on the data.

Matthew: Is there’s a little bit of moral hazard where people, for whatever reason, change the way their property is defined?

Mark: That’s a good question. How do we avoid the moral hazard if somebody’s deliberately re-coding? Just like moral hazards in car quotes, when people say they park in the garage and they see the price change, there are interactions with customers which change the opinion about their home.

We have multiple sources to check against. They say it’s a cottage, but I can see a picture of it and it’s a block of flats and I know Barclays lent the mortgage to them. We’re looking for subtle inflexions. One thing is affecting price and the other is delivering a good user experience. We can do both without necessarily challenging the moral hazard.

Matthew: How do insurers connect to the WhenFresh database?

Mark: Either through APIs, a combination of API and flat files, or just flat files. For us, it’s more important that the data is right than it being fast. It needs to be well considered and prudent, so we’re helping insurers reduce loss ratios.

Providing the right information doesn’t always mean having to send it out by API, sometimes it means giving an insurer a file to run locally. That’s good, that’s not a problem. We roll with whatever people are using and if it’s a legacy system, we’ll work around it.

Matthew: And you offer test access to your data via an API to let people try before they buy. How does that work in practice?

Mark: It’s usually done in two stages. The first is identifying the data the insurer needs. Typically, they might say, ‘I have 1,000 policies and I’ve suffered a loss on some of them. Can your team append them with everything WhenFresh knows?’ We’ll work with their analyst or lend them an analyst to identify common factors to avoid or price differently.

Then we either provide access to the API or a flat file of all of the releases in the UK where that ingredient is true. They can test it on the sandbox to see in real-time if the thing they don’t like is present in a candidate seeking insurance.

Matthew: There’s a big move towards reducing the number of questions customers have to answer to take out insurance. How confident can insurers be that the data that comes back is accurate?

Mark: There’s a new bank using a hook up to our API where they aren’t sending out any physical evaluations or surveys. It means there isn’t someone doing a virtual visit to collect all the other data, questions around Japanese knotweed, condition, plot size, restrictions. But we do and working with our partner CLS we’re able to merge the data sets to produce an insurance wrapped data block. That means our customers can take out loans in three hours. Not just the approval, they could actually deposit the cash.

They don’t need to send out a valuer to look, they don’t need to send out a survey. They can hit our combined APIs and check 75 different fields on the shape of the roof, quality of windows, presence of Japanese knotweed, proximity to water. The insurance wrapper also covers the conveyancer, so the collector doesn’t need to run those checks. They can go straight to lend.

Matthew: So, you’re offering the data for valuations and an insurance product covering incorrect valuations?

Mark: There are two products. One is called Verify, which judges whether the value of something is correct. If a bank lends on it and they have to repossess the property because it turns out the value was wrong, the insurance covers that risk. As a result of Covid-19, the real problem is not the valuation but the condition. We work closely with Hometrack to provide an insurance policy that sits behind the valuation. That’s not the difficult bit, the difficult bit was no longer being able to send a surveyor.

Any insurance company not using prefill is missing a trick. There’s no point in asking somebody a question we already know the answer to. Going back to moral hazards, there’s no point in asking my mother how tall the nearest tree is. She’s 82, she wouldn’t know. We’ve got the satellite imagery, so we know how big the tree is. But are we willing to risk a mortgage based on that data? Not only are we sure, we’re so sure we’ll insure it. We’ve moved out of just selling to insurers into using data to build insurance products for insurers.

Matthew: We’ve been talking about building valuation, but from the insurance perspective, it’s the rebuild costs that are important. Are you also tracking that?

Mark: Yes we are. One of the rebuild costs that often gets overlooked, but we’ve found a really good way of doing it, is alternative accommodation. I was talking to an insurer yesterday morning about the challenges around exactly that. There are models for handling flood and remediation in the event of flood damage, but the cost of rehousing somebody during repairs is part of that. Insurers need a map of all the things that are rentable, that approximate to all the things insured.

That’s what we’re getting into with the rebuild cost. It’s not just the bricks, the cost of local builders or dealing with the local authority. It’s also about what happens to the customer during the work.

Matthew: UK insurers are less demanding on policyholders to define the rebuild costs. Does that increase the value of what you’re offering to clients?

Mark: The challenge we see is insurers using a current market valuation of a property as a proxy. It’s true except in circumstances where it isn’t. Imagine a £100,000 mid-terrace unit that is catastrophically damaged. There could be £200,000 worth of damage if it’s sitting in between two other houses.

Frankly, the value of the thing is also not the same if it’s reduced to rubble. Its value is not minus, it could be minus some number of hundreds of thousands of pounds. We need to know the value of those things and figure it out from there.

Matthew: What can you tell us about the work WhenFresh is doing with the Bank of England?

Mark: We are looking at changes in valuation over populous. What things are out there that are buildings and what are they worth? The BoE uses that data in their macroprudential reports and it’s not a secret that the report that came out in April quoted analysis we had done. That’s what we’re doing at a high level. It’s a bit more than that, but that’s their IP, not mine.

Matthew: Is there anything else that we should be keeping an eye out for from WhenFresh in the next few months?

Mark: I would look out for our combination of data and insurance as a product. The valuation is taken off the AVPS, which is what we use to insure a building so companies can lend on it. There’s more stuff coming to watch.

Also, I’m looking forward to seeing people again. The people we met at InsTech London events were always phenomenal. I hope we get back to that safely as soon as possible. I miss our Tuesday evenings.

Matthew: One final question, where does the WhenFresh name come from?

Mark: To find that out, people need to listen to your podcast!