(Image credit: Adobe Stock.)
In two previous articles published at the Novarica blog, I wrote about the three distinct phases through which insurers progress in their efforts to implement big data technologies and work with the Internet of Things. Though different insurers have achieved varying levels of maturity for each technology, one key lesson is evident: there is real value available for a company even at earlier stages. These two technologies are strongly linked with a third, emerging technology: artificial intelligence, or AI. AI will enable insurers to derive new insights out of the immense quantities of data captured via IoT devices and stored and manipulated within a big data environment.
AI technologies are still in their infancy; no insurer can be said to have reached “maturity” in that space, making it difficult to talk about phases of AI implementation. But it is still possible to think about different approaches to leveraging AI, and insurers should consider which values are most important to them before beginning any AI initiative.
For these approaches, we’re looking at AI for insight, meaning consuming and analyzing data to advance business goals. (As opposed to leveraging chatbot and natural language processing technologies for better customer interactions.)
Approach 1: Better Insight… or: With the same amount of data, make better decisions.
The Better Insight approach utilizes AI to augment human decision-making with the goal of better results over time. Using AI to mine existing data for new or better business insight is the value most people imagine when considering how these technologies will impact the industry. And this is certainly a key area where insurers can look to improve their business. Can AI help underwriters make better decisions about incoming risks? Can AI help claims adjusters make more accurate reserve estimates? Can AI do a better job identifying fraud?
A key element of this approach is understanding what we mean by the “same amount of data.” This isn’t referring to the databases of existing policies and claims history, though, yes, any AI approach will be relying on those warehouses to feed its machine learning processes. That existing data is essentially the AI process’s experience, just like human underwriters and claims adjusters bring their own experience to the table. Rather, the “same amount of data” means the incoming data that requires a decision, e.g., the policy submission or the first notice of loss. AI leveraged for better insight should be able to help make better decisions using the same incoming data that a human would use.
Approach 2: Better Automation… or: With the same amount of data, make decisions in less time.
The Better Automation approach utilizes AI to speed up or fully automate a business process with the goal of maintaining parity of results. Rather than focusing on better decisions, this approach focuses on faster decisions. The question an organization needs to answer for this approach is whether it can leverage AI technologies to take humans out of the decision-making process entirely without suffering a decline in quality of results. The goal in this case is to provide real-time responsiveness to agents and policyholders. For some insurers, the goal may also be an ability to do more with less human staff.
Once again, in this approach, the assumption is that the same amount of data is available to any AI elements as it would have been to a human decision-maker. Here AI is replacing the human instead of helping the human make better decisions.
This automation vs. augmentation dichotomy aligns well with Professor Tom Davenport’s view of AI’s future, a hopeful vision of task automation rather than job automation and of the evolution of knowledge work into a more augmented profession. But there’s also a third approach, one particularly suited to the insurance industry, which is continually struggling to build a better customer experience for its complex product space.
Approach 3: Better Experience… or: With less data, make the same decisions.
The Better Experience approach utilizes AI to make decisions without requiring as much data gathered by agents or policyholders. Insurers have always embraced the value of data-driven decision-making. In recent years, however, there has been a push for better customer experience, faster turnaround times, and mobile applications that take up less space. This has created competing needs: more data gathered to support better decision-making and less data gathered to support simpler, faster interactions.
Unlike the previous two approaches to AI, this third one allows for a reduced set of data at the entry point for a decision—for example, a photo or two of a damaged vehicle rather than a long write-up after an auto accident; a shorter questionnaire during a policy application process; or the ability to underwrite large life insurance policies without requiring medical tests. In these cases, AI may have the ability to both mine an insurer’s long history of experience in ways humans cannot and leverage new sources of IoT data that are too dense for human investigation, resulting in an ability to make decisions even with less input from the source.
Combining the Three Approaches
Together, these approaches look at three key factors in decision-making: the QUALITY of a decision, the TIME needed to make a decision, and the INPUT that informs the decision. With AI entering the equation, there’s also the long-term goal of combining all three approaches: with less data, make better decisions in less time.
It’s unlikely that AI solutions will be able to swoop in and change all three fundamental factors of decision-making at an insurance company. Insurers will want time to validate results and correlate them with future losses, view AI-based and human decisions side by side to compare outcomes, and experiment with different levels of data input to see if loss ratios can be maintained.
These approaches—unlike the phases of maturity for big data and Internet of Things—aren’t tiered, meaning different approaches might be tried for different areas of the business. For an enterprising insurer, this might mean simultaneous attempts to augment underwriting decisions, automate fraud detection, and simplify the data requested during the FNOL process.