Machine learning and AI-driven pricing has huge significance for the insurance and risk management world – and drone use is just the beginning
How do you work out the price of insurance for a fleet of drones all flying at different times, in different places, with different wind conditions, some flown by experienced pilots, others by rank amateurs?
It sounds impossible.
But live pricing has become a reality for advanced insurance buyers, and drones are just the start.
Flock is a London-based InsurTech which underwrites on behalf of Allianz Global Corporate & Speciality.
Flock’s CEO is 26-year-old entrepreneur Ed Leon Klinger, who tells Insider Quarterly that artificial intelligence (AI) makes it possible for the company to rapidly interpret huge amounts of information.
“We take in 50,000 flights of data, split them up and quantify the risk of every one of those flights,” he says.
Since January 2018, Flock has been gathering data on around 1,500 commercial drone users that have used the InsurTech’s app to buy cover.
If a pilot wants to fly on a windy day, their insurance will cost more, while a flight at rush hour over a road will have a more expensive premium than the same flight a few hours earlier.
“It’s not just the data. It’s the interpretation of the data,” Klinger explains.
The company has pulled the views of risk assessors into its pricing model. Flock also has its own claims experience to work with, having been selling insurance on its app for more than a year.
“At Flock we believe the technology we’ve built has massive implications outside the drone industry,” Klinger adds.
Although the entrepreneur refuses to be drawn on other lines of business the startup is exploring, general aviation insurance is one clear area where real-time, pay-as-you-fly insurance can make its mark.
At its core is the power InsurTech can bring to solving a very real problem facing insurance: data quality.
But the implications of AI-driven insurance pricing ripple out into the wider waters of how companies manage risk.
At the heart of changes happening in the insurance industry is the idea of enriching the underwriting process with useful information pulled from outside sources.
Underwriting using external data doesn’t just make life easier for the insurer, it makes life easier for the client.
As CEO of Munich Re Digital Partners Andrew Rear explains: “One of the most frustrating things about insurance is you have to answer all these questions.
“The great thing about using external data is we don’t irritate the customer.”
Azur, an AIG-backed InsurTech that underwrites UK high net worth insurance, is a good example of the benefits of using external data.
Once a broker has pumped a customer’s address into Azur’s new pricing system, the company knows the age of the property, its value, and even any planning applications the owner has submitted to their local council.
Small business cover
For the US small and medium-sized enterprise market, Munich Re-supported InsurTechs are scraping data from sources like Yelp and TrustPilot.
SME startup Next Insurance, part of the Munich Re Digital Partners network, has run “a whole bunch of experiments” using external data to price risk, says Rear.
Munich Re’s traditional reinsurance business is also doing tests with incumbent insurers that cede risk to the reinsurer.
According to Rear, it is often easier to feed new data sources into InsurTech startups than with existing insurers.
“These integrations with modern data systems are quite easy to do, with legacy systems they are quite hard to do,” he explained.
For high-volume, low-value lines of business like personal lines and SME commercial, it makes a huge amount of sense to pull in external data to aid underwriting.
Meet the robots
For external data to be useful, it needs to be interpreted properly.
This is where artificial intelligence comes in.
AI refers to any attempt to simulate human-like intelligence, encompassing everything from a robot that can play football to a chat bot which can pass the Turing test and appear human.
In data science, AI can teach itself how to interrogate huge stacks of data.
Machine learning is a branch of AI which leverages existing statistical models. Artificial Neural Networks were first proposed in the 1960s and, with today’s computing power (specifically, the very same hardware used to run computer games) can be used to “learn” the patterns behind massive data sets.
Machine learning isn’t really new. As Charlie Blackburn, chief technology officer of AIG-backed InsurTech Azur puts it: “Machine learning is really just 25-year-old Bayesian mathematics”.
Londoner Thomas Bayes was a Londoner who died in 1761 and is buried next to Silicon Roundabout, the area around Old Street station in the UK capital which is popular with tech startups. He gives his name to the branch of statistics that seeks to find answers in oceans of data.
Although mathematicians have long theorised that it is possible to find predictive patterns in vast data sets, it is the advance of cloud computing over the past five years that has given data scientists more processing power at their fingertips than ever before.
In some lines, like direct motor, machine learning is already being used to price risk. For others, it is only a matter of time.
“In almost all lines, machine learning is going to transform underwriting,” Rear notes. “We’re already using it in small business.”
Indeed, machine learning is already widely used across the insurance industry in highly commodified lines like UK motor.
Digital Fineprint, a London-based InsurTech, is getting rather good at scraping and processing information from the internet. The company scoops up online data sources like the UK’s Companies House, social media and online reviews to help insurance companies price risk and brokers find new sales leads.
The InsurTech’s founder Erik Abrahamsson launched the company out of business school, having previously worked at Twitter.
“We read reviews about businesses, we take in sentiments analytics and compare it to [insurers’] internal data. We allow underwriters to become more like a partner,” he says – in many cases giving them more information than the client would have access to themselves.
Abrahamsson’s vision of the insurance industry is one where the underwriting role itself is increasingly taken over by actuaries and data scientists building pricing software.
The role of an underwriter then becomes much more about working with clients to manage risk.
Moving up the value chain
But machine learning and AI are no longer just helping retail and SME insurers. The technology is rapidly moving up the value chain to help mid-market commercial insurance and wholesale and specialty cover.
Richard Hartley is co-founder and CEO of Cytora, an AI-driven InsurTech that has taught its models on high-frequency business lines like property insurance for restaurants, working with the likes of Starr Companies and QBE.
He says Cytora’s machine learning processes are getting better through time, the more data they ingest.
The startup sells a tool to underwriters that scans broker submissions, ranking them by the quality of the risks, with analysis by a cornucopia of metrics – from the state of a firm’s financials to the distance to the nearest fire station.
The likes of Cytora, Digital Fineprint, Omnius and Carpe Data all offer routes out of the current crisis in underwriting, acquisition costs and expenses facing the industry.
Taking Lloyd’s as a microcosm of the commercial and specialty scene, the market’s combined ratio has worsened over the last seven years – a period in which deterioration in non-cat claims was flattered by an astonishingly benign period for cat events.
Meanwhile, a company like AIG is grappling with the consequences of its “large limits” policy for the firm’s profitability.
Insurers worldwide are changing the way they underwrite. Carriers are looking closely at the way brokers are paid amid scrutiny from regulators. Meanwhile, the back office is an area where costs have to come down if the industry is to recover profitability, as global warming threatens to increase cat claims.
Technology can help address all these challenges.
The marine market, for all its glorious ink-stained tradition, has been one of the worst culprits for embracing technological change.
The Lloyd’s mariners mutinied two years ago, when electronic placing was first introduced in the London market.
But under pressure from clients, some senior figures in the marine insurance market have embraced change.
A marine blockchain initiative by Maersk and EY is now supporting more 500,000 transactions on the blockchain, insuring 1,000 ships.
AI technology is being deployed by marine InsurTechs like Windward and Concirrus. Both firms are applying machine learning to publicly available data sets, as well as from completely fresh sources such as satellite data.
So should brokers and underwriters fear the rise of the robots?
Almost every technology entrepreneur quizzed for this story believes the role of the human underwriter remains crucial for higher-value risks.
Azur CEO Graham Elliott talks about the future insurance market as a world of “augmented underwriting.”
Other startups, including the MGA C-Quence, are working on bringing analytics based on machine learning to the mid-market.
Both companies are targeted brokers – the very group with the most to fear from AI.
Azur operates in the high net worth market, while C-Quence writes on Arch paper in mid-market UK commercial lines.
In theory, if commercial insurance can be automatically underwritten by an algorithm, brokers and underwriters will be redundant – quite literally.
At lower premium levels, startups like Next Insurance, and incumbents led by carriers like Hiscox, are proving the direct model can work. But in the mid-market, InsurTechs are so far focusing on making life easier for buyers, brokers and underwriters.
For Hartley, Cytora is all about giving underwriters more time and information to underwrite, helping them spend less time trawling through broker submissions.
For others, like Azur’s Elliott, the broker is ultimately the one who understands their client best, and is in the prime position to cross-sell across multiple lines.
For example, the web platform built by Elliott’s team allows a professional lines broker, perhaps visiting a client, to see if the director buying cover for their business wants to add on personal home and cyber cover for themselves. An address is pumped into an app, and two minutes later a quote is produced.
Like most technology, the client doesn’t actually care how the product works behind the scenes.
For canny brokers, these tools will help cement existing relationships and provide opportunities to push into new markets.
Now for the big stuff. For reinsurers, AI can help carriers understand what it is they are actually insuring.
“Reinsurers just get incredibly poor information from their cedants,” notes Hartley, who adds that Cytora delves into reinsurer’s portfolios to monitor for large losses.
“Even if you’re a reinsurer, you can still get the same information that a broker would [using the software].”
For example, he says, an insurer can find out every fire or flood claim in the UK on the system.
Using advanced machine learning and AI, once the data is aggregated, it is as possible for a retrocessionaire or a reinsurer to discover exactly what their risk is, building by building.
Geospatial Insight is another company that is using technology to help reinsurers. The InsurTech pulls images from satellites, light aircraft and drones to monitor risk around the world.
When the Fundão tailings dam collapsed in Brazil in January, with the loss of 169 lives and the destruction of around 100 buildings, Geospatial Insight was able to rapidly inform clients about what was happening.
“The imagery was supplied first, followed by the analytics, including an assessment of the potential damage or destruction of properties in the immediate region,” a spokesperson told sister publication The Insurance Insider.
What companies like Geospatial Insight can do is use AI to monitor aerial photographs for signs of change at insured sites. “From a pre-risk perspective, underwriters can access enriched information on which to price and assess property risk.”
The approach is especially useful for reinsurers, where risks are bundled up and further disseminated across the market into retro and facultative placements.
AI digital monitoring of a location can give carriers critical information about the changes to individual risks being underwritten within those portfolios.
“Continuous monitoring helps prevent disasters, such as the Vale [tailings] dam, as assets are evaluated regularly, using automated change detection techniques to identify asset degradation or signs of damage,” the company said.
Geospatial Insight argues that, even before the Fundão Dam collapse, the structure was being undermined, with leaks visible from aerial imagery.
“Insurers are using this technology to better understand risks, take steps to prevent losses, provide loss estimates, support claims and inform future modelling.”
For Adrian Jones, deputy CEO of P&C partners and head of strategy at Scor, there are now countless new sources of data to underwrite and manage risk.
“You can use this data to have a more intelligent conversation with the client about your client’s risks,” Jones says.
But, he warns, AI is “only as good as the data coming in”.
So if robot-analysed external data is to truly transform the industry, it must start at the point of purchase and not be degraded along the way.
Once the technology being used by organisations like PPL, Whitespace, Digital Fineprint, Azur and C-Quence is commonplace in distribution and placing, (re)insurers can then start delivering real change to the industry.
Jones paints a future of insurance as one where the retro underwriter (who, for the sake of artistic licence, we will put on a sun-lounger on a beach in Bermuda) will have access to “the same high-quality risk data, scraped from all available public sources, as the retail broker in an office in Springfield”.
There will be enough information for risk be properly traded across the value chain, making it easier for third-party capital to underwrite primary risks.
And then the robots can really come out to play.
This article was first published in the Spring 2019 issue of Insider Quarterly.