Yup, AI Really Is Everywhere

By: Dr. Gary Anderberg

By: Dr. Gary Anderberg

May 16, 2023 — What about AI and claims handling? AI, in one form or another, seems to be turning up everywhere, so why not in claims? Well, some folks at Stanford and MIT have already been working on the idea, so let's take a look at what they have come up with thus far. Their research paper was published by the National Bureau of Economic Research (Generative AI at Work | NBER), but it's available through Cornell University without charge if you're resourceful.

Before we plunge into the findings, let me note that there are two flavors of AI that are commonly referred to interchangeably. This research focuses on AI using machine learning plus large language models (LLM). This is not quite the same AI as our Waypoint reserving system here at GB, which is designed to understand behavior patterns instead of language and is not as specifically suited to the type of work examined below. Technical points aside, the following suggests several ways in which an LLM+AI may be applicable — in a few years — to claims and related client-facing services.

Here's the basic research setup: "We [the authors] study the staggered introduction of a generative AI-based conversational assistant using data from 5,179 customer support agents." The idea — which we think highly of — is not to have the AI do the job but rather to provide expert assistance to the people doing the task to allow them to complete their work sooner and with fewer errors. This is a type of decision support that is not disruptive of normal client — customer relationships. It provides what the authors call "conversational guidance for customer support agents."*

While the study sports a dazzling array of charts and graphs, the conclusions are pretty simple. "Our paper provides the first empirical evidence on the effects of a generative AI tool in a real-world workplace. In our setting, we find that access to AI-generated recommendations increases worker productivity, improves customer sentiment, and is associated with reductions in employee turnover." That sounds pretty good to us. The model operation used in the study is not exactly the same as a claims shop, but it's close enough that the same approach appears viable.

My friend, Peter Rousmaniere, has studied the new report in detail. Here's his take on the application to insurance claims, a subject he knows well:

• First, the authors study the problem resolution rate of service workers. They find that low-skilled workers improve their problem resolution rate relatively quickly compared to those unaided by AI. The productivity improvement is estimated at 10 — 20 percent. Their main point is that low-skilled workers move more quickly into higher performance rates.
• Second, AI brings low-skilled workers the tacit knowledge of higher-skilled workers — knowledge that is hard to include in written training guides and brief training sessions.
• Third, AI softens the tone of communications between service workers and customers, which eases up tensions that often arise.

Will AI be turning up on every street corner next week? Not likely, but some small portion of the AI hype appears to be real. The study showing that a well-designed AI can help lower-skilled workers** move more quickly into higher levels of performance is the real gold. All developed economies are in the same demographic squeeze with us older, deeply experienced, been-there-done-that types retiring in droves.***

The type of AI described in this study may also be thought of as extraordinarily efficient on-the-job training. Making new employees (all employees really) more effective and less error-prone, note, can have a serious positive impact on E&O exposures. Properly understood, AI may be the most effective risk control tool now on the market.**** Finally — a real win-win for all of us.

*GB's Waypoint application is a good example of this approach to decision support, instead of leaving the work to the AI and hoping for the best. As the results noted here show, people and AI working together are a potent combination.

**Adjuster trainees, for example.

***Well, not all of us. Some of us are enjoying the party too much to leave.

****This is your real takeaway, friends. While AI poses risks (see below), it can also be a powerful risk management tool. Look for the yield; ignore the hype.