‘You need to be sure that you can trust the system, in particular in claims,’ says chief product officer
Agentic artificial intelligence (AI) has the potential to move firms beyond minor productivity gains and into the realm of full-cycle business process automations that could make major improvements to balance sheets.
This is according to Rob Malan, associate partner at McKinsey, speaking as part of a panel at AI insurance platform Shift Technology’s agentic AI webinar.
While generative AI tools such as chatbots can create new text or images based on user inputs, agentic AI goes a step further, giving the AI “agent” the capability to interact with its environment in an executive manner and the autonomy to perform certain tasks.
Malan explained: “The wave of generative AI has been ramping up, but in most cases, it’s still having single-digit [productivity percentage gains].
“Most players have struggled to get real, tangible improvements from generative AI, aside from some marginal productivity gains and interesting experimentation on the side.
“But what agentic AI promises is true end-to-end intelligent automation of complex processes that will really hit the expense ratio in a very powerful way, while improving outcomes for customers and employees.”
Claims handling
Claims handling has been identified by many in the industry as a prime candidate for the embedding of agentic AI workflows.
Read: Offering configurable agentic AI marks a ‘paradigm shift’ – Earnix
Read: One in three SMEs have no form of cyber insurance
Explore more artificial intelligence related stories here, or discover more news here
While each claim is different and nuanced, many steps in the process involve repetitious manual work, something which agentic AI proponents say can be well automated – especially if working in tandem with a human adjudicator.
According to Eric Sibony, chief science and product officer at Shift, reliability and explainability are the keys to achieving buy-in for the technology.
He said: “Reliability remains the key challenge if you want to have fully automated processes. You need to be sure that you can trust the system, in particular in claims. You wouldn’t want to [use automation] if you didn’t have an accuracy above 99%, because any error in claims is so impactful.
“When it’s about an insurance process, if [the AI agent] gives explanations about the data points it used and the rules it’s using, a claim handler can assess if they’re doing the right thing.”

He graduated in 2017 from the University of Manchester with a degree in Geology. He spent the first part of his career working in consulting and tech, spending time at Citibank as a data analyst, before working as an analytics engineer with clients in the retail, technology, manufacturing and financial services sectors.View full Profile
No comments yet