Back-office functions face the greatest automation exposure, but insurers risk customer trust if AI deployment weakens oversight in claims and underwriting.
Insurers are accelerating artificial intelligence adoption across back-office operations, but greater automation could increase risks around fairness, human oversight, and customer trust.
Drew Propson, Head of Technology and Innovation in Financial Services at the World Economic Forum, said data-heavy back-office roles are amongst those most exposed to AI.
Across financial services, four out of five areas where AI is heavily deployed involve back-office functions, she said. Customer service is the notable exception, with financial institutions already seeing benefits from automation, including higher customer satisfaction.
Roles requiring relationships, experience, and complex judgement are less likely to be displaced. Propson pointed to sales as an area where human interaction remains important, whilst experienced employees will still be needed for complex underwriting and claims decisions.
As insurers automate more decision-making, data quality will be critical to maintaining fairness. Propson said training data should accurately reflect the populations to which AI models are applied, whilst insurers can conduct pre-deployment testing to identify potential biases.
Human oversight will remain equally important, particularly when AI encounters unusual cases.
“You want to make sure that any time that there’s an edge case or an anomaly in the decision that an AI might come out with, you’re able to have some extra oversight and have another pair of human eyes on it,” Propson said.
Moving too quickly could also create longer-term commercial risks.
Propson warned that customers are likely to notice failures quickly when insurers deploy AI at scale without sufficient safeguards, potentially damaging trust that can be difficult to rebuild.
As adoption accelerates, customer experience could therefore become a key competitive differentiator, forcing insurers to balance automation gains against the need for reliable decisions and human judgement.