Artificial intelligence will help workers across industries offload simple, repetitive tasks to automation, freeing employees up to spend more time-solving problems that require a more thoughtful approach. But how should you determine which decisions can be made by A.I. and which still need a human in the loop?
It's a question of particular importance to the insurance industry, where agents receive thousands of claims every hour but can only address so many in a day. To lessen the load, mega-insurer Aflac has begun using A.I. to approve "simple" claims, and has built several guardrails to determine which decisions the A.I. can and can't make.
Keith Farley, Aflac's senior vice president of individual voluntary benefits and former director of innovation, says the company introduced an A.I. platform last year that can approve claims that don't require proof-of-loss, such as adding a dependent to your plan or paying out wellness benefits for getting regular checkups and screenings. To ensure that the A.I. couldn't make any major mistakes, the team programmed it to only approve claims. Any recommendations for denials are automatically surfaced for human review.
Farley acknowledges it's possible that the A.I. could over-approve some simple claims that might have been denied by a human, but the time being saved through the automation outweighs the cost of those approved claims. With that time saved, Farley says insurance agents can devote more time and energy to helping claimants who are processing intense emotions like grief and need to be treated with empathy.
So, how can other businesses draw a line for what tasks A.I. should or shouldn't be used to handle? Farley says it all comes down to the task's margin for error. For example: Aflac recently launched a generative A.I.-powered application portal for individuals looking to find employment at Aflac. "If somebody uses our application portal to make an appointment for 3pm on Tuesday but it's actually at 2pm, that's not ideal but I can handle it," says Farley.
When it comes to claims, though, Farley says you have to be right 100 percent of the time. "You need to be able to explain why you made the decision that you made, so there's really no room for error. If I pay a claim wrong because generative A.I. made a decision and I can't find where the data came from, that's unacceptable."
No comments:
Post a Comment