A 2024 McKinsey study (“Now Decides Next: Insights from the Leading Edge of Generative AI Adoption”) found that 79% of executives across a range of businesses expect “substantial transformation” from AI “over the next three years.”
These business leaders understand how AI can accelerate critical functions, such as underwriting and claims operations in insurance. However, betting on specifics around fast-moving tech carries substantial risks, including the tech “aging out” before deployment.
Many recognize this challenge.
“We need to think about what tasks are manual, repetitive and routine; but also cognitive, drawing inferences and interacting emotionally,” Dr. Anand Rao, Distinguished Service Professor of Applied Data Science & Al at Carnegie Mellon University, explained in a recent Roots Automation webinar panel discussion. “That’s going to help us understand which ones to automate – and which to augment.”
Alex Taylor, Global Head of Emerging Technology at QBE Ventures, offered a persuasive example: “Understanding the text within documents, not just extracting it, is the magic,” he explained. “In some cases, we’re looking at hundreds of pages explaining a risk, which could be three or four hours that an experienced underwriter is spending just to get started – time not spent pricing a policy or picking up more submissions. The opportunity to accelerate [their productivity] is profound.”
“Little Strokes Fell Great Oaks” –A Case for Incrementalism in AI
So, what approach to implementing AI works best – and in which situations?
“Start by focusing on simple use cases – summarization, extraction, classification – in high-volume, low-complexity cases,” said Ratish Dalvi, VP of Al & Machine Learning at Roots Automation.
“Automate smaller workflows, or augment one aspect of decision-making and build from there.” Dalvi added, “try seeing if simple rules can solve a problem first—use patterns to fix a process programmatically. But when the number of rules you need to apply gets into the thousands, they’re changing over time, and the variations in the data require regular interventions—you need to look at AI.”
“Augmentation is more important towards the mid- and high-level commercial insurance, for example,” said Dr. Anand Rao.
In situations with changing rules, contexts or information, “you really want the human judgment from experienced underwriters.” And when they’re augmented by a context-sensitive model trained on insurance data – like InsurGPT™ – the effect is massive.
“We had a small team of underwriters dealing with a huge number of submissions, thousands of pages of documents,” says QBE’s Alex Taylor. “They could never assess them all. Using AI models to extract the relevant information, underwriters were able to go from looking at 10% of submissions to 100% almost overnight. And they welcomed [AI-augmentation] because it meant they were doing the bits of the job they actually wanted to do.”
Upskilling and Retraining: It’s Not Just for People Anymore
This brings us back to the very premise of AI in insurance: empowering human capability.
“Don’t spend too much time trying to automate tasks that require a lot of judgment,” warns Dr Anand Rao. “It’s a classic Pareto rule: automating 80% of the tasks is relatively easy, but the last 20% is enormously challenging.”
Tools like Roots’ AI-powered Digital Coworkers are just that – coworkers, part of a team. “An AI on an automated retraining schedule collects data and works it through human-in-the-loop processes,” says Ratish Dalvi. “That’s how you make the models resilient to changing patterns in your business. That regular retraining is key to avoiding ‘model drift.’”
The ideal system will learn from your best people to augment and raise human performance across the board: “We know AI very quickly gets entry-level people up to average performance by taking the judgment of your senior people into the process,” adds Dr. Rao. And then, over time, the average performance rises, too.
Automate the simple. Augment your people’s decisions with the right information. And accelerate its delivery. “AI is faster – the value is not that it’s more intelligent… at least not yet,” adds Ratish Dalvi. “It’s the speed, especially around its ability to learn, adapt and scale.”
The takeaway: The more you automate, the better and faster you augment.
Hear more from Dr. Anand Rao, Alex Taylor and Ratish Dalvi on how to move your organization from AI hype into AI-powered hyper-productivity.
Watch the Roots Automation webinar, “Harnessing AI Document Processing Across Underwriting & Claims,” on demand for expert insights and guidance on how insurance businesses are accelerating processes with generative AI tools and a look ahead at the next five years of insurance’s AI transformation.