Artificial Intelligence
Insurance works with large amounts of data, about many individuals, many instances requiring insurance, and many factors involved in solving the claims. To add to the complexity, not all insurance is alike. Life insurance and automobile insurance are not (as far as I know) the same thing. There are many similar processes, but data and numerous flows can be different. Machine learning (ML) is being applied to multiple aspects of insurance practice.
Insurance is about risk. The insurance industry sets rates based on expected payouts so that, hopefully, they end up with positive revenue. Setting rates and understanding payout in order to maintain profitability is complex, and the industry hope is that ML can help in achieving that goal. Note, here, Im focusing more on ML than artificial intelligence (AI), because many of the complex statistical tools that are now considered ML can more efficiently accomplish some of the tasks than would neural networks, expert systems, or other purely AI tools.
There are multiple ways machine learning can help in the insurance industry. Let us take a look at three.
Health and life insurance are complex. There are multiple factors that go into understanding an individuals risk factors for disease, illness, and mortality. Insurance underwriters have historically used a core set of factors such as male/female, age, and smoker/non-smoker. When other factors have been used, such as zip code, the problem of red-lining has appeared in insurance as well as the more well-known area of financial red-lining. Therefore, there are regulations about how some demographic information must be used.
The need to address those legal concerns means that underwriting isnt only about individual health risks, but about legal risks as well. Analysis must be done to back out some features that could cause legal risk while still creating pools that remain profitable.
This is where machine learning comes in. The performance of modern computing can handle large amounts of data, and complex regression analysis can perform clustering that can help with analysis. Those ML techniques provide value without the need for AI. For insurance underwriting, statistical models and procedural code are providing an improvement in analysis for companies, said Paul Ford, CEO & Co-Founder, Traffk. We are working with neural network models, but the overhead for training and runtime must be balanced with the accuracy improvement necessary to make it worthwhile to roll-out those engines. While things might change, down the road, our existing ML models provide advances in analysis and profitability for our customers.
At the other end of the insurance process is the issue of claims. It is not only the insured who have problems with claims complexity. In the automotive industry, the need to understand the variety of repair options and parts available create a challenge for both service providers and the insurers.
With automotive claims, providing an estimate based on the typical costs for repair is not sufficient. Its not only that vehicle types vary, within a class of vehicle the repair costs can vary based on the insurance coverage, as well as the availability of parts in geographic regions.
Machine learning can help with claims in a number of ways. In addition, multiple ML tools can be used throughout the claims process.
Take the First Notice of Loss (FNOL), the initial notification to the insurer about the accident or damage. If theres a quick estimate of total loss, theres a different process flow that is much simpler. No ML is needed in the review of damage, but robotic process automation (RPA) might be used to simplify the claim flow to payment.
With other damage, or even to understand if there is a total loss, ML can be used. The most obvious tool is AI vision, but even this can have multiple processes. A phone app can step a customer through taking pictures that an AI system can then analyze for damage, with a backend AI system working to link to parts and estimate. A repair shop, in comparison to the insured, is more familiar with the process and can have a different front-end asking more detailed questions to more quickly get a more educated response from the repair experts.
Note that two different approaches were mentioned. It would be overly complex to have a single AI system that could support every step in the claims process. More efficiency is gained by letting separate systems process claims, identify damage and provide repair estimates, said Evan Davies, CTO, Solera. By using different approaches to machine learning through the claims process, you maximize the benefits of automation and enable skilled workers to focus on more complex cases.
One thing Evan Davies also pointed out was how the process flow can change depending on the severity of accident or the type of insurance coverage provided. Minor damage and standard coverage can be fully automated, as all parties are fairly comfortable with the process and dollar amounts. Totals, as mentioned, dont require AI. Those claims in the middle, however, can be helped with an adjuster reviewing the analysis and working with the customer, for the benefit of both short term monetary issues and long term customer relations.
Yes, we keep coming back to fraud. Sadly, it is a human condition and a risk in so many areas of business. Insurance is no exception. As Ive recently talked about fraud and ML in other business arenas, I wont go into detail here. Let it be sufficient to point out that analysis of claims doesnt stop at processing all claims as if they are proper.
Cluster analysis is used to understand, for instance, if a similar type of accident is happening in an area at above normal amounts; potentially indicating organized fraud.
In the analysis of potential fraud, multiple tools are used, some are in ML, such as statistics, rules based approached and even neural networks.
See the rest here:
Machine Learning Can Help The Insurance Industry Throughout The Process Lifecycle - Forbes