Machine learning: The next step for insurance analytics
Some industry observers categorize the insurance industry as in the middle of an arms race when it comes to data and analytics. Carriers are competing to get the best insights about customers so they aren’t stuck with only the worst risks.
But after insurers are done with the first level of investments in data and analytics infrastructure and staff, a new front will open up, according to Novarica’s Jeff Goldberg, SVP of research and technology. The coming battleground will be in using machine learning and artificial intelligence to get even deeper insights about policyholders and prospects.
Jeff Goldberg, Novarica“If your competitors aren’t doing it you’re fine – but as soon as your competitor starts using a different tool or model, you start suffering adverse selection,” he explains. “We’ve been going through that with predictive analytics, and we’ll get to the point where we are with machine learning and AI too.”
Goldberg recently co-authored the “Artificial Intelligence in Insurance: Application and Use Cases” brief for the insurance technology analyst firm. AI represents the next level of maturity with data and analytics, he says, with wide-ranging implications for insurance technology strategy.
“We’re not yet at the point where we’re talking about truly intelligent machines,” he says. “We’re talking about the ability to mine data at a level we’re not able to right now, with computers able to make cognitive leaps about data value.”
As an example, Goldberg cited the case of Google’s translation algorithm. Google taught it to translate from English into Korean, for example, but not Korean into Japanese. However, because the computer was taught Japanese as well as English and Korean, it was able to translate between Korean and Japanese by creating its own internal language to which all phrases were applied before output. That’s what separates AI and machine learning from traditional analytics, Goldberg explains.
“Taking cognitive leaps – that’s kind of what your goal is with data analytics with machine learning,” he says. “We give it this first level, let’s see where it goes.”
But the march toward AI doesn’t mean insurers need to scrap their existing plans for big data and analytics. There are several years of incremental steps in between the state of things now and full automation, and there are plenty of insights and advantages to gain over the next several years, Goldberg says.
“From a strategy perspective, you can’t stop caring about iterative improvements of data analytics just because there’s a possible future where everything is automated by machine learning,” he concludes. “We’re going to see stages along the way. It’s certainly not 50 years off, but it’s not one year off either.”
For example, Goldberg says, many insurtech startups (and tech companies in other industries) are looking for both big data experts and machine learning specialists, so that they can work in concert to produce the best results.
““It won’t replace your data scientists, it will inform them,” he says. “When you’re hiring someone who knows how to build predictive models and build insight, you will also have someone who knows how to program a deep learning environment.”
Goldberg's co-authors on the report were SVP Jeff Benton and associates Stephanie Dalwin and Jon Leslie.