Artificial Intelligence for Insurance: Friend or Foe?
Over the past five years, we have witnessed substantial startup activity across a broad range of financial services sectors, with fintech funding overall surging from $2.2B in 2011 to $14.4B in 2015 (per CB Insights).
Until recently, one of the sub-sectors of financial services that has been somewhat outside of this trend has been property & casualty insurance, which is a multi-trillion global industry and growing. That is now beginning to change as VC investments in insuretech (technology based solutions for insurance) has grown from ~$130M in 2011 to ~$1.7B by the end of 2015 (per CB Insights). Building on this momentum, Oliver Wyman sponsored the inaugural InsureTech Connect conference (October 4th – 6th in Las Vegas) bringing together industry incumbents, innovators, and investors to discuss how new technologies — ranging from big data and online marketplaces to drones and artificial intelligence — may reshape the insurance industry.
At Omidyar Technology Ventures, we are particularly excited about the potential impact of artificial intelligence (“AI”) on insurance. During InsureTech Connect, we moderated a panel of entrepreneurs who are pioneering the use of AI in insurance and wanted to share the insights they offered about the short and long-term business opportunities and the paths to broad adoption of AI in insurance. The panel consisted of Adam Cassady at Tyche, Chris Cheatham at Risk Genius, Chris Augeri at Drive Spotter, and Adithya Raghunathan at Captricity.
In the short term, we believe AI will achieve the greatest uptake by solving problems for industry incumbents and addressing areas that offer a high and rapid payback and also fit neatly into existing workflows. At present, we are less clear on the near-term opportunity to use AI to displace industry incumbents. (For example, while there is some enthusiasm for the potential of general AI powered businesses to replace insurance agents in the near-term, we believe that the best opportunity for AI today is to work hand-in-hand with brokers and agents and to empower them to improve their efficiency and customer service.) Within the AI value chain, we see opportunity in the following areas:
Marketing, customer acquisition, account management:
The abundance of data available on both individuals and businesses can be used to target customer segments more effectively, provide more targeted recommendations once customers are in the sales funnel, and enable more personalized service. For example, Captricity helps insurers better understand their customers by using AI to convert unstructured data from handwritten and faxed documents into structured data that can then be mined to enable improved customer insights and service. Meanwhile Risk Genius uses machine learning to quickly compare commercial policy options for customers and identify potential gaps in coverage in a fraction of the hours it might otherwise take an agent or broker to do the same task manually.
With the wealth of public data available on people, organizations, and the world (e.g., online identities, weather, news), AI can enable more fine grained risk segmentation, reveal otherwise hidden risk factors, and identify previously unknown correlations between combinations of risk factors and claims outcomes. For example, analysis of individual Facebook profiles may offer a far more accurate determination of somebody’s health choices than simply relying on the insurance applicant’s representations; mining Twitter feeds may offer insight into how well a business is being run and hence how likely they are to be taking the appropriate precautions in managing both property and potential liability risk exposures. For example, Carpe Data has worked with insurers (who have shared their historic underwriting files and claims outcomes) to develop a range of predictive algorithms based largely on publicly available data, including social media (by the way, Carpe Data has a pretty good idea of whether or not you’re a smoker and their assessment doesn’t rely on finding a “smoking gun” picture of you on Facebook). Cape Analytics uses computer vision and machine learning to extract insights from geospatial data to provide an underwriter with essential attributes about a homeowner’s residence, such as whether the roof is due for replacement or repair. Meanwhile, Tyche is analyzing both structured and unstructured underwriting and claims data to unearth powerful insights into the drivers of claims activity. For example, do you think that a painting company that uses scaffolding rather than ladders is a higher or lower insurance risk within the North-east? Within the Southwest? And how much better or worse is it? Tyche knows!
Beside generating underwriting insights, AI can be used to help advise on how insurance customers can take actions that mitigate risk so that underwriters are willing to reduce their premiums. For example, Drive Spotter applies data analytics and machine learning to a combination of telematics data and streaming video captured from cameras on board commercial vehicles to help fleet managers monitor and improve driver performance. But reduced premiums are just one layer of benefit realized by their fleet manager clients. Drive Spotter also yields huge returns through insights that help to meaningfully reduce both fuel consumption and maintenance costs.
Moreover, AI can add value simply by ensuring that underwriting rules and standards are more consistently applied. In certain areas of insurance human bias still plays a big role in underwriting decisions and inconsistencies in how similar policies are underwritten by different people within the same organization can be surprisingly large and meaningful.
Fraudulent claims cost the insurance industry $30B annually. By integrating more data into the claims review process adjusters can better assess the validity of a claim. For example, someone who just put in a claim for a lost or stolen item might also be sulking on Facebook about all the money they lost gambling. Probably a good claim to consider for further review. Shift Technology helps insurers review claims and identify those that are likely valid and should be paid immediately, as well as those that raise questions and would benefit from further investigation.
Additionally, the claims process is long and tedious and often heavily manual. We’re excited by how Tractable uses deep learning to remake the automobile repair estimating process. Tractable’s AI is able to instantly assess the damage severity to a vehicle just based on photos, optimising repair and claim management processes. Simply have the driver snap a few pictures after an accident and Tractable will analyze the damage instantly and enable a much faster settlement.
The good news for all these startups is that a number of large incumbent insurance providers (e.g., AIG, AXA) are already experimenting with AI. That said, selling a novel AI technology solution into an industry that is not a conventional early adopter will not be easy. It will be essential that startups offer a clear value proposition, fit neatly into existing workflows, understand the data available (and their ability to make use of it without a substantial and time consuming data preparation), and are realistic about the enterprise sales cycle. Because the current application of AI requires data that is ideally structured/labeled data and offers sufficient volume, veracity, velocity, and variety (the “Four V’s”), it will also be important to screen for opportunities based on an assessment of the customer’s data availability and readiness. Moreover, some insurance companies and data providers may be hesitant to release the data or the data might just be bad. In such cases, companies might consider creating their own data as an alternative, or perhaps finding creative ways to access and scrub data more efficiently. That said, this selling hurdle may come down over time — while today’s data hurdle for training algorithms may be quite high, emerging AI solutions hold the promise of generating valuable insights on much less information.
We’ve highlighted where we’re enthusiastic about the near-term value creation for AI in insurance, though what really gets us excited is the long term opportunity for AI to come together with real-time data obtained through IoT sensors to reinvent the insurance industry. We believe we’re heading toward a future where insurers look increasingly like risk managers that work in partnership with customers to reduce claims exposures/activity for their shared benefit. Imagine a future world where your home is completely connected (e.g., Google Home, Canary) and communicates with your insurer. You receive a phone call notifying you that your water consumption appears much higher than usual and to check for leaks. You find a broken pipe in your basement though fortunately you catch the problem before there is too much damage. You are also offered a premium reduction if you implement an automatic water shut off valve (this is already being done today). Or perhaps you receive another urgent call asking you to check your wiring as the monitoring of your electrical activity (the biorhythm of your home’s energy usage) suggests you have a faulty electrical line. You find the faulty electrical line and realize that you potentially avoided a fire that may have consumed all of your possessions and cherished memories.
We’ve long admired how this prevention-oriented insurance model has been deployed in the large and mid-size commercial account market to great success by companies such as FMGlobal and Hartford Steam Boiler. While historically this model has relied heavily on on-site engineers, we see a future where AI + IoT enable this compelling business model to be sufficiently automated so that it can scale down to the smaller commercial and even consumer customers thereby creating substantial value for insurers, customers, and society at large.
We would love to hear from others who are thinking about insuretech or from startups tackling tough industry problems with an AI angle. We’d also love to hear any thoughts or reactions to our blog, so please feel free to reach out to Chris or Pearl. Additionally, we wanted to say a special thanks to many of the people mentioned in the blog for sharing their feedback and insights.
Authors: Chris Bishko, Pearl Chan • October 27, 2016