Episode #45 — Mark Varley the CEO and founder of AddressCloud reminds us that geo-coding is not a solved problem and explains why and how inaccuracies during the geocoding process can have consequences in terms of risk assessment models used by insurance companies. Mark walks us thought how and why his company built its own geo-coder and why locating and describing addresses with rooftop level accuracy is the first step in building risk profiles. We also discuss the changing role of geo and GIS in the insurance industry.
This episode is sponsored by HiveMapper
A platform that takes video and creates 3D mapping layers based on that data. The video can be from a variety of different sensors, does not need to be vertically looking down on the geography and each 3D output is georeferenced!
You are more than welcome to reach out to me on social media, I would love to hear from you!
In Conversation
An Accidental Geographer in Insurance
Daniel: Hi Mark, welcome to the podcast. I’m really pleased you could take the time to come along and do this interview with me today. In the pre-interview we talked a lot about risk assessment. I know that you’re involved with insurance, and obviously you have a company called AddressCloud, and we’re slowly but surely going to put that whole story together for the listeners. But right at the start, perhaps you could give us a brief background of your history. How did you get involved in geospatial, and what does geospatial have to do with insurance?
Mark: Thanks ever so much for the opportunity to come on and speak — big fan of the podcast, listen to lots of episodes and it’s really refreshing, so I really appreciate the invitation. My background — I’m another one of these accidental geographers, I think you’ve had a few on the podcast. I graduated in 2000 and started my career, as probably a lot of people do, on one of these grad schemes. So I went and joined Accenture, for no reason other than they were paying the most at the time — that’s I guess what attracted me into the consulting industry. Didn’t really know where I was going to end up, and got placed in the financial services practice. I did a brief stint with the London Stock Exchange and then I found myself working in insurance. This would have been around 2003.
Mark: Again, no geo particularly — I was working in the general consulting practice. But bizarrely, which I didn’t really realize at the time, insurance has got a huge location element to it. At the time we were engaged in a big project to basically take GIS, which had been a background function for many years — a very specialist part of the insurance business — and really bring that front and center of the client’s operations. What they were looking to do was take location and looking up addresses, apply some GIS, and have that happen as part of the quote and buy journey. So when a customer was getting their insurance, really focusing in on that address level and being able to give a really good accurate view of risk and price. That was my introduction into GIS, and I’ve been learning ever since.
Geo With No Maps
Daniel: Could you give us a little bit more background on that introduction to GIS? There was obviously an understanding in the industry that location was really important in terms of assessing risk. How detailed were you getting with that location — was it a street level address, was it a postcode, was it city limits? And if you could put some words around what kind of data you were using to make those assessments, that’d be really interesting for the listeners.
Mark: Absolutely. Historically the insurers have been using geo — they’ve been some of the quite early adopters of geo, probably going back certainly to the 80s and potentially even before that. But a lot of the work that was happening was desktop-based, and it tended to be the very large commercial risks, where there’d be one or two really smart people who’d know about GIS, who probably had a couple of ArcGIS licences, or whatever the precursor was. They would go and look up risks that looked kind of big or scary. The big risks in the UK tend to be flood risk and fire and arson risk — we used to get a lot of home fires.
Mark: Historically most of this kind of process was happening what we’d call post-event, or post-binding, so often the insurer was already on cover. And anything that was happening at the point of quotation was really at the postal code level. In the UK we’re blessed with quite a detailed postcode system, so we can get you down to perhaps a city block, maybe somewhere between 50 and 100 properties. What we were doing as part of this work was really saying, well, let’s take it down to the next level, let’s get it to building and actually to address level.
Mark: The classic example there is where you’ve got a street that’s on a hill, and at the bottom of the hill you’ve got a river. If you’ve got somebody who’s living at the bottom of the hill, they’re close to the river, they could be a worse flood risk than potentially even their neighbour who may be only two or three doors away. That’s a very different risk profile. So what we were doing was really trying to use geography and geo techniques as part of the quote process. We almost called it “geo with no maps”, because although maps and geo work were really important as part of this particular process, the client would never see a map — but maps were being used in the background to inform the price and eligibility.
What Goes Into a Risk Model
Daniel: That sounds like a really big leap — that understanding that within postcodes we can have these different hazards, these different risks, and then we can assess each house individually. So it sounds like you’re becoming a lot more granular in your approach. Can you give us an idea of what other kinds of data sources you might use? I can definitely see it with the example of elevation and being close to a river, but what other kinds of risk factors might you take into consideration?
Mark: Absolutely. At the time, the company I was working for, we would buy in a lot. We had an in-house team who were modelling things like crime — taking police statistics and Office of National Statistics census data — so they were modelling out crime. The flood risk we were actually bringing in from a third-party company, and that was taking in elevation data, DEM elevation data, and they actually simulated rainfall as well. The company we worked with was a company called JBA, who were the market leader in the UK for flood risk, and they actually have products that go around the globe — they’re busy trying to do world domination at the moment. They’ve got some really clever software that basically models terrain and then simulates water flow, and understands where that will build up, where rivers potentially would burst, as well as surface water and coastal overtopping.
Mark: So it does very much flood and crime, but there’s definitely some fire risk in there as well — things like taking building outline information, dissolving building outlines, working out where fire could spread, and simulating those kinds of events as well.
Daniel: It’s worth noting this is probably not a one-time calculation either. I’m assuming flood risk will change depending on how the urban environment develops, and also in terms of fire — how is the landscape changing, how are the buildings changing, what are the socio-economic factors within a certain postcode? I’m assuming these things need to be recalculated, the model needs to be rerun from time to time. Can you give us an idea of the frequency of that?
Mark: Absolutely. The flood work is done by — we don’t do that ourselves, we work with a third-party company who do that, but they’ve probably got 100 people who are working on that on a daily basis doing edits. So taking into account, for example, where new roads have been laid, where the environment’s changing, when new houses have been built. The data is normally refreshed annually, so flood data is refreshed annually, but they’re doing updates and edits on that data daily essentially. So it’s an ever-changing, complex beast.
Why Geocoding Isn’t a Solved Problem
Daniel: Now that we’ve talked about geospatial and risk assessment and the insurance industry, I think we understand why it’s important and how it’s used. Can you talk a little bit about what AddressCloud, your company, is doing, and how you are improving the situation, or what problems you are solving?
Mark: Sure. I started the company five years ago. It was just me initially, and it was me for the first couple of years. We’ve now grown — we’re still a pretty small company, there’s about five of us now — but really we set about trying to help specifically around geocoding. The problem we had at the time was that the company I was working for, we’d invested a huge amount of money in these third-party datasets. We were taking data from the Ordnance Survey. In the UK we’re blessed with a really good postal code and addressing system, but we’re slightly held back by the fact that a lot of the data is closed source commercial data. So to be able to get to rooftop level with geocoding requires quite a large investment in data, through essentially government public data. But then it also requires a good system to be able to match the addresses and to do that at scale, and we were really struggling.
Mark: We’d invested in this fantastic data that was taking postal office data, local authority data, and bringing that into a single product. But the system we were using to search that data was pretty cumbersome — it was an in-house application that we had to host, and getting data updates was really challenging. For example, if you’re an insurer you need to have really up-to-date information. If a property’s just been built and you want to price it at address level, you need to know where that address is straight away and be confident that it’s in the right place. That was a real challenge for us as a company.
Mark: I was looking at different solutions out there, I thought there was definitely space for a new one, so I decided to quit my job and essentially build what has now become AddressCloud. The idea was to really have a kind of Google-like experience — really simple to use, nice interface, really easy to integrate with, fully cloud hosted, fully managed and in the cloud. And that’s where AddressCloud was born. Basically the name does what it says on the tin.
Daniel: In terms of geocoding, isn’t that problem being solved? There are lots of different geocoders out there — how are you guys different? I think we understand now how geocoding relates to and is a really important part of risk assessment in terms of insurance, but that bit that you guys are doing — you’re doing your geocoding and then you’re enriching the data, is that correct? And then somehow the insurance company is getting hold of that and giving quotes based on your data. Can you fill in the gaps a little bit there for me?
Mark: Absolutely. As I say, we started very much as a pure play geocoding solution — initially that’s all we were doing. It’s one of those things that people think is a solved problem until they come up against it. There are some good general purpose geocoding solutions out there — people like Google and Bing and HERE, for example, are doing a fantastic job on a global scale. But when you get into within a country, you really need to have specialist knowledge of the colloquialisms and the way people refer to addresses, which is very much a country by country specific problem. So that’s where we thought we could differentiate, by having something that was really trained in on the UK and Ireland, and being able to do that job using government data, and being able to do that to something where an enterprise would be able to use it and get that extra degree of accuracy.
Mark: What we found really is that once we’d done that — I wouldn’t say it’s a solved problem, it’s always a challenge, there are always edge cases and bizarre things that fall into the “you can’t make it up” camp — but once those started to die down, our customers were then saying, well, this is great, you can tell us where this address is and we’re really confident in your results. What else can you tell us about that address? So that’s really been our focus over the last two or three years: bringing in as many high quality datasets as we can, really linking those all back to the address, and having that available as an API and a service that is very quick, very scalable and reliable.
Daniel: That’s really interesting, what you said about geocoding, because I have largely seen this as a solved problem — we have the solution, we have many different solutions. So it’s really interesting to hear you talk about it. One of the most interesting things about it is it’s such a critical part of geospatial. If we can’t take our human understanding of addresses, a human way of geocoding, and translate that into something that’s spatially enabled that the machines can understand, we can’t connect all these different data sources. So it’s a really important thing, and the fact that it’s not 100% solved, that we can’t yet with 100% certainty say we can solve this problem every single time — I think that’s really interesting.
Mark: And I think as part of that as well — if you think about us, and I’m putting myself in the GIS person camp there, which I still feel like a bit of an imposter about even after all these years — there’s one thing around geocoding, which is finding yourself on the map and getting your bearings and getting yourself hopefully in the right location. But when you take the map away, and when you’re working with something like this where everything’s happening in the backend and it’s also informing ultimately whether or not people are going to get insurance and what they’re going to pay for it, you need to be really confident in that location. I think that’s where the real challenge lies — reducing those false positives and making sure that you’re returning an accurate result.
Daniel: Absolutely. This is not a visual thing, there’s nobody looking at this. This is a point-in-polygon lookup — okay, I have a property here, I want to insure it, and then you go to your system and say what are the risks in this immediate area, what is the risk profile here.
When You Can’t Get to the Address
Daniel: While we’re talking about risk profiles, I’m assuming sometimes there must be some edge cases where you can’t locate a property, or you have some troubles around it anyway. Can you then look at the risk profile of adjoining properties, properties that are close by, and then assume a certain amount of risk based on that? And do you look out into the future, do you do forecasting as well, or is it just a here and now risk assessment that your software delivers?
Mark: That’s a great question. What we do is we model the hierarchy of a property. Both in our geocoder and also in our intelligence service, we model the hierarchy and recognize that we can’t always get to address level, particularly where there’s not a user present. So often our customers will take, let’s say, a big portfolio of addresses from a broker. Brokers in the UK specifically seem to have a real issue with address quality — some are better than others, but we get some really bizarre addresses that have got typos and things all over the place. We do our best to get them to address level, but we can’t always get there.
Mark: So what we do is we offer the customer the option to drill back. We wouldn’t normally go to an adjoining property. What we would normally do — when we say address, we’re going kind of beyond building. So we’ll give different scores, for example within an apartment block we’ll know if the property is on the ground floor or the first floor or on the top floor. From a flood insurance perspective that’ll be a different risk profile. You might not want to insure a ground floor property for flood, but you might be quite happy to do that for the property that’s on the second floor above.
Mark: Where we can’t get to address point level, we can drill back to building level or potentially to postal code level. But the key thing really is being confident and flagging up that level of accuracy to the insurer. And then obviously depending upon the risk — if it was a penthouse apartment in a really salubrious part of London that was worth an awful lot of money, and we couldn’t get them to address level, could only get them to postcode level, they might want to then trigger a referral process to go and check that on a map. Or for a low value property they might just be okay — that’s fine, we’ll go ahead, we’ve got a postcode level risk and we’ll write that anyway. So that approach differs very much depending on the insurer and on their risk appetite.
Daniel: That was a really interesting observation — that of course the risk profile is going to change, if we think about flooding, depending where the property is located, if it’s on the third floor, the fourth floor. I often fall into that trap of still thinking in 2D, but obviously we live in a three-dimensional world, so that was a really important point.
Accumulations and the World Trade Center
Daniel: In the pre-interview you talked about something called accumulations, and I found this fascinating. Would you mind describing that for the listeners?
Mark: Absolutely. Rewinding back to why insurers started investing in geography — ultimately an insurer’s biggest cost would be their reinsurance. Typically insurers in the past have kind of almost been writing blind. If you imagine you get these big insurance groups where they might own many different brands and different subsidiaries, often all writing and sometimes competing with one another, they would often not know historically if they were all insuring in the same building.
Mark: The classic case — very tragic, but classic case for this — was the World Trade Center. When the 9/11 event happened, the customer I was working with at the time, they owned many subsidiaries who were all insuring in both the twin towers and the surrounding retail, and it almost ended up being a catastrophic event for the insurer. It almost brought the insurer down, just because they were essentially writing in one place. That’s what we call accumulation management. It’s basically understanding where you’re insuring and, do you have too much in one space? Is it too much that could potentially introduce a risk to the company? So accumulation management is really about the process of having good quality data, cleansing that data, plotting it on a map, and then looking to see where your hot spots are — where do you have too much risk, and how can you potentially mitigate that risk?
Daniel: Thanks for taking the time to share that. I’d like to dive into these risk profiles a little bit more and try to understand what it is that people are getting out of the system. Do they get a separate risk profile for things like flooding, fire, crime, or these other sorts of hazards? Or is it just an overall weighted risk profile?
Mark: There are a couple of other companies who work in our space, and different people take this in a different way. What we tend to do is we give a full breakdown. For any given particular address we may have anywhere up to sort of 100 or 200 different attributes, and we’ll break that down by the different kinds of risk — as opposed to coming up with a sort of black box magic score, you know, write it or don’t write it. So what we typically deliver back is a very detailed assessment. We try and avoid generally giving an opinion. It’s more: okay, here are a bunch of things that you need to take into account. And then the insurer would plug that into their own algorithm to work out a price for the customer.
The Future: Real Time, Climate and Geo Moving Front and Centre
Daniel: This was probably a little bit unfair of me, but at some stage there I asked two questions and we didn’t get an answer for the second one, so I’m going to try again. The question was: is this a here and now risk assessment, or do you look out into the future? Is there some kind of modelling that you do to say, we assume that this risk will change in such a way over the next five or six years? When I think of this I’m thinking in terms of urban development, for example — we know a long way out in the future what’s going to happen in a general area, we have plans, and that could change the flooding profile.
Mark: For insurers, insurers typically think in 12-month cycles. Most insurance policies are normally valid for 12 months, and then in 12 months’ time we renew and we reevaluate. So ours is very much at the moment focused on here and now. But it is a great question. What we’re seeing is some of our data providers and partners, and we ourselves, are being asked now about taking into account things like climate change, as other industries start to adopt these services. We do predominantly insurance but not exclusively — think for example about lenders. A mortgage provider or a home loans provider is essentially taking a 25 or 30-year gamble on a property, and they’re wanting to really take into account that kind of information. Climate change data: is the risk profile of this property that I’m essentially loaning the money for my customer to buy going to be the same in 10, 20, 30 years’ time? So it’s not something that we do much of at the moment, but definitely something we’re being asked about.
Daniel: I’d like to move off now and talk about the future, because I’d be really interested to hear your thoughts on what this kind of risk assessment might look like in the future. Are there any trends you can see happening, any movements in any one particular direction in the industry at the moment?
Mark: I think we see a few things happening — not necessarily completely in our space, but a few general trends that I think will affect us as a business and also the industry. As I mentioned, in the past GIS was very much a back office function. It has moved now and often is present in the front for many insurers, but I still don’t think it’s reached full scale adoption. So I think there are going to be more and more insurers who are going to be doing this as part of their journey. And I think really GIS is going to be moving — I’m even seeing that structurally. Some of our customers, we would often work with GIS, and a lot of insurers have got their own in-house GIS departments, and a lot of that now is becoming merged and really just seen as a more specialist branch of business intelligence, as opposed to a very specific function. So I think geo is becoming really front and center, and the insurers are recognizing the value of it.
Mark: And then in terms of the IT side — I’ve worked in the past for a big consultancy where there’d often be these big three or four year engagements, big projects to put in these huge systems, and that’s kind of going away now. A lot of insurers tend to be going out and purchasing specialist services, third-party APIs, and they recognize that while it’s very important, it’s a very specialist area. So we see that happening more and more often.
Mark: Insurance generally is still quite a manual process, particularly large commercial corporate insurance. We work with some customers who trade through the Lloyd’s market, and if you imagine the classic stock exchanges back in the 80s, people waving around bits of paper — it’s not quite the same in Lloyd’s, but it is still very much paper-driven. You’ll go into the City of London and you’ll see brokers walking around still with these great big portfolios and lots and lots of paper. I think over the next 10 to 20 years that’s going to go away, and we’re going to get to a state a bit like the stock exchanges were in the 80s and 90s — they all became automated and everything went electronic. I think the same is going to be happening for the insurance industry as well.
Daniel: It was really interesting to hear you talk about the role of GIS and geospatial professionals in the industry, and how they’re moving slowly but surely from the back office up to where the decisions are being made, and even being viewed as part of business intelligence. That’s something we’re seeing across other industries as well — more of a recognition that hey, this is maybe not as special as we used to think it was, but it’s definitely data, and it’s a really important part of these models and it’s going to be a part of making more granular, precise decisions.
Daniel: In terms of insurance, what we’re talking about here is obviously making better decisions — we hear that term everywhere — but we’re also talking about spreading the risk. We talked briefly before about these shorter life cycles of an insurance contract. Do you think we’re going to see that increase and become ever smaller and smaller, until we get to a stage where perhaps we’re assessing risk in real time?
Mark: I think so. I think there’s definitely going to be more real-time information coming in. The nature of what we do, we tend to be working with property insurance, and that tends to not be necessarily so important. But if you take into account other classes of risk — so for things like marine risk, tracking ships, there are heaps of interesting use cases where actually having that real-time element becomes really important.
Mark: One use case that our software typically gets used for, and something we are increasingly working on and being asked for, is around post-event analysis. Understanding, where a big event has happened — for an insurance company a couple of things they need to know pretty much straight away is what’s their potential loss, what’s their potential exposure to that event. That could be a big flood event, a big fire, could be a terrorism event. They’ll need to know very quickly what their exposure is, and they would often use GIS to be able to do that. And then also some of the insurers actually had a department where if there’s been a big flood or a big event like that, they would actually send out a mobile claims unit to go out there and help the customers, or potentially make outbound calls to the customer to say, are you okay, how are you affected, is there anything we can do to help? Again, having that real-time information and being able to intercept those, I think is really important.
Daniel: In a previous episode I talked to the CEO and founder of a company called Watchkeeper —
Mark: I heard that one. It was a great episode, really good.
Daniel: What I thought was really fascinating was: not only are they assessing risk — it was in a different vertical — but assessing risk in real time and making decisions based on that. But the idea that it could then become a model, and they could go back in time and watch the risk accumulate and these factors play into the end result over three, four, five days. I couldn’t help but think, when you were talking about where this risk assessment is going in this platform that you’ve built, if that’s not what you’re building as well — if you couldn’t wind back in time or fast forward in time at some stage, and be able to see the risk perhaps not on such a granular level for each house, but for postcodes or larger geographic areas, and see how it changes over time and what factors are really affecting that level of risk.
Mark: I think that’s great. As I say, we haven’t seen much demand from customers around having that sort of temporal view, but you could certainly envisage a solution where that was playing out, and that would be really super interesting. We do start archiving — we have got an archive of what we call post-event boundaries and footprints, and over time it would be super interesting to have a look and see how those have evolved and if there are any patterns.
Mark: And again, playing back onto your previous point about geo becoming more and more recognized and valued as part of the core business function — obviously we’ve seen the last couple of years, some geo people, if they’re smart, are reimagining themselves as data scientists. But then you also get data science people coming in who haven’t got a geo background, who are then discovering geo. I think as all that converges, having those two quite different disciplines with different academic backgrounds and different ways of viewing a problem, I think we might start seeing those kinds of things happening more and more.
Daniel: Mark, on behalf of the listeners I really want to thank you for coming along and sharing your insights, sharing your knowledge, and telling us a little bit about the risk assessment industry and the insurance industry. I’ve really enjoyed the conversation. If people want to reach out to you, what’s the best way for them to do that?
Mark: Thanks ever so much again for the invitation. You can contact us on our website, addresscloud.com. I’m on LinkedIn — Mark Varley, V-A-R-L-E-Y. We are on Twitter, we don’t tweet a lot, but @addresscloud, you can find us on Twitter as well.
Daniel: Thanks Mark, I really appreciate it.
