Insurance executives who wonder if artificial intelligence should be a top priority for their company might want to consider the following:
- “AI is one of the most important things that humanity is working on. It’s more profound than, I don’t know, electricity or fire.” – Google CEO Sundar Pichai in a comment to MSNBC and Recode earlier this year.
- “We’re rapidly headed toward digital super intelligence that far exceeds any human. I think that’s very obvious.” – Tesla and SpaceX CEO Elon Musk in an interview in the new documentary “Do You Trust This Computer?”
- A recent PwC study of 2,500 U.S. consumers and business decision makers last year found that 72% business executives believe AI is going to be a fundamental business advantage.
While the insurance industry has already taken a proactive approach to AI and is using the technology in a number of important pilot projects, given its likely impact there’s much more to be done. Here are three strategic steps insurers should take to prepare for an AI tidal wave:
Focus on “process improvement,” not complete automation. As important as it is, AI is not a panacea for all business challenges. Machine learning algorithms can be part of the solution—but by themselves won’t solve the problem. Police departments, for example, are using Amazon’s facial recognition service to improve the process of trying to match an artist’s rendition of a suspect with the department’s mugshot database. The top five most likely mugshot matches provided by the service include the correct matchup 70% to 80% of the time. While not a fully automated solution, this approach represents a huge improvement over having to manually scan thousands of mugshots to identify a match.
Provide clean, actionable data. When it comes to the quality of the data that they rely on, AI and machine learning are no different from every other computer program: put garbage in, get garbage out. To provide good results, AI applications require consistent and accurate data that is error free. To ensure that their data is up to snuff, here are three questions insurers should ask themselves:
- Do they have enough data on a specific topic or concern—fraudulent claims for instance—for an AI program to identify meaningful patterns?
- Does their data include the key attributes that are needed—credit scores or criminal histories, for example—to reach the desired conclusions?
- Is their data accurately tagged? Are pictures of damaged homes, for example, properly labeled for use by claims processing?
Unless these questions can be answered in the affirmative, any AI or machine learning initiative will be a nonstarter.
Focus on the industry’s most likely use cases. With machine learning, it’s important not to try and boil the ocean all at once. While a wide range of use cases are certainly possible, insurers should focus first on the most likely applications. These include fraud detection, improved customer service, support for adjusters and pricing conversion modeling.
Using the example of fraud scoring, once an insurer has the proper data in place to support a machine-learning algorithm, it can institute thresholds for which scores should be sent to straight-through processing versus standard claims processing or the insurer’s special investigations unit (SIU).
This accomplishes three things: It improves the efficiency of the insurer’s adjusters by reducing the number of claims they need to process; it improves the customer experience, since legitimate claims can be paid more quickly; and it helps SIU concentrate on claims that are most likely fraudulent and avoid wasting limited resources investigating legitimate claims.
In a similar fashion, machine learning can help expedite image processing, reducing the paperwork currently inundating many insurers.
As AI adoption increases, the onus is on insurance executives to determine which areas of their business the technology can benefit the most. If they don’t, their competitors surely will at their expense.