Alternative data involves more than just numbers for insurers
The insurance industry's data windfall not only bestowed an incredible bounty, it also raised red flags.
Insurance has always been a data-driven business — just ask any actuary or underwriter. However, the incredible volume of personal information becoming available through sensors, third-party aggregators, and other alternative sources, as well as the evolving technology tools to make sense of it all, have not only bestowed an incredible bounty on the industry but are raising red flags as well.
A running theme throughout Deloitte’s “2020 Insurance Outlook: Insurers adapt to grow in a volatile economy“ is how insurers should take analytics to the next level and use alternative data sources to broaden engagement with digital consumers. The outlook also emphasizes the flip side, looking at potential pitfalls of expanded data access and analysis, including heightened cybersecurity and privacy considerations.
Opportunities and challenges involving analytics were front and center at a recent symposium on “Data in the New: Transforming Insurance.” There was a mixture of hype and caution expressed during the event, which was run by the Tobin Center for Executive Education at St. John’s University. A wide range of speakers from the worlds of insurance, InsurTech, and academia spent the day sorting through new products, processes, and business models that may be enabled by alternative data, along with possible speed bumps likely to be raised by regulators and privacy advocates.
It’s about quality, not quantity
With so much real time information in hand and far more on the way as sensors proliferate in vehicles, homes, commercial buildings, machinery, and wearables, insurers trying to make sense of it all may feel like they’re trying to drink from a fire hose. Yet more is not necessarily better. Indeed, some at the event warned that insurers could drown in a flood of data if they’re not careful about how they parse and utilize all this new information.
For example, some of the academics questioned the return on investment in auto telematics. How much does it cost to collect and leverage all the data being gathered about driving behavior versus the relative value received? How are loss probability and pricing models being affected? What’s the outcome of underwriting and pricing decisions based on telematics versus the use of traditional proxy data, such as age, type of vehicle, or credit score? More research likely needs to be conducted before insurers can confidently assess whether telematics is more predictive — and if so, whether it is more predictive enough to make a significant difference in their bottom line.
A few of the speakers even wondered whether debating traditional versus alternative auto insurance data is worthwhile given the ongoing experimentation with autonomous cars. Should driverless vehicles become the rule rather than the exception somewhere down the road, that could eventually render not just telematics but traditional auto policies obsolete by taking drivers out of the equation, moving the risk largely into the realm of product and professional liability underwriters. Given such uncertainty, how much should insurers be investing in collecting and analyzing such data in the short term?
Of course, sensors are driving more than just auto insurance — with data via embedded devices supplemented by alternative information collected from social media, third-party aggregators, and even drones. The biggest question will likely not be accessibility, but utilization. How might artificial intelligence (AI) and advanced analytics be deployed to generate actionable insights and allow insurers to monetize all this raw material? How could enhanced connectivity and alternative data help reinvent conventional insurance policies and influence product development?
Privacy concerns on the rise
Meanwhile, insurers should expect an intensifying privacy debate as consumers and lawmakers challenge assumptions and conclusions based on alternative data and/or how it is collected. For example, if a consumer indicates on a life insurance application that they do not smoke, is it okay for the insurer to check social media to see if there is a picture of them smoking, or use voice recognition programs to check for evidence of smoke-related vocal damage? How about confirming an applicant’s weight — should they be allowed to apply facial recognition technology to determine the prospect’s body mass index?
On the property-casualty side, is it all right for a workers’ compensation insurer to fly a drone over a claimant’s home to see if they are really incapacitated, perhaps catching them mowing the lawn, doing roof repairs, or swimming in the backyard pool? As insurers have access to more personal information from a growing variety of sources, potentially thorny ethical questions such as these will likely need to be sorted through.
Data and analytics validity may be questioned
The validity of alternative data and analytical tools to leverage it may also be challenged. Regulators will likely want a peek into an insurer’s “black box” to examine the algorithms or AI prompting alternative data-based decisions on underwriting, pricing, and claims. Are automated systems valid, fair, and unbiased? One could argue that the human actuaries, underwriters, and claims adjusters handling such decisions with traditional methods may have similar issues to address. However, while you may not be able to argue with an intelligent machine or automated algorithm, regulators can still hold the people using them at insurance companies accountable.
Those at the symposium discussed a three-part test that insurers may soon face to assess the validity of alternative data and related processing programs. How open and able they are in responding could make the difference in winning over skeptical regulators and customers:
- Is there an actuarial basis for what the insurer concluded from its alternative data and analytical programs? What is the correlation? How valid are the underlying assumptions? How predictive is the result?
- Is there a difference between an actuary’s or underwriter’s opinion and an AI-generated decision? What about professional judgment? Is there a sound and reasonable justification showing that alternative data has a causal link to the risk being assessed? Is there any human oversight or regular audits of machine-driven determinations? Can AI-based decisions be appealed?
- How easily can AI determinations be defended? Is the accuracy of alternative data independently validated — particularly when purchased from third parties? And might there be any disparate impact, even unintentional, affecting any particular group?
Full disclosure could make a difference
Even if all the new data being collected is found to be reliable and the analytical systems employed deemed fair and reasonable, there remains the more fundamental question of how much disclosure insurers should be required (or choose) to make. As more alternative data become actionable via advanced analytics and AI, insurers will likely be called upon to provide defensible explanations.
A Deloitte research paper on privacy released earlier this year suggested insurers and other financial services firms should be more transparent about how they collect data, why they need it, and, most importantly, how sharing it might benefit not just the company but the consumer. Insurers may be able to discourage resistance, avoid conflicts, and perhaps even differentiate by treating alternative data as a tradable asset rather than just a compliance consideration relegated to mandatory privacy statements that most people likely never read — or if they do, may not understand. Proactive engagement could set the stage for improved customer experience, greater service options, and more cost-effective solutions.
There are several potential advantages for insurers, individually and as an industry, in clearing all these hurdles. Alternative data and advanced analytical tools could enable a host of game-changing transformations in how insurance is marketed, distributed, priced, and administered. Among the possibilities:
- Greater customization to support the gig and sharing economies, with carriers determining via telematics what kind of insurance a buyer needs, and when.
- Rethinking a policy’s value proposition by turning insurance (where carriers pay for damages after the fact) into assurance (where they help policyholders prevent losses from happening, to their mutual benefit).
- Enabling real-time support systems, such as providing health and fitness advice as well as services in conjunction with a life insurance policy, or safety monitors built into employee clothing by workers’ compensation carriers.
- Automatically-triggered parametric insurance based not on a claim being filed, but an event taking place, or a threshold reached.
Unfortunately, insurance is still too often an undifferentiated product sold to a generally indifferent or reluctant consumer. That could change if carriers play their cards right with alternative data, AI, and advanced analytics by making insurance a more dynamic, individualized, and customer-driven service.
Click here for a copy of Deloitte’s “2020 Insurance Outlook.” You may also click here to listen to an archived version of Deloitte’s Jan. 8, 2020, outlook webcast.
Former National Underwriter Editor in Chief Sam J. Friedman (samfriedman@deloitte.com) is insurance research leader at Deloitte’s Center for Financial Services in New York. Follow Sam on Twitter at @SamOnInsurance, as well as on LinkedIn.
These opinions are the author’s own. This piece is published with permission from Deloitte. See www.deloitte.com/about to learn more about Deloitte’s global network of member firms.
Also by this contributor: