Baringa warns insurers’ AI ambitions outpace results

Human staff would still handle judgement calls oversight and exceptions.

Insurers are increasing their use of artificial intelligence (AI), but many are still struggling to turn early projects into business-wide results.

The main barriers are not the AI tools themselves, according to Baringa’s “Reimagining the insurer in the age of AI” report. 

“The obstacles are rarely about the technology itself. Instead, legacy systems, fragmented data and organisational constraints are slowing progress and preventing promising innovations from scaling,” it said.

Baringa said a small group of insurers has already completed large-scale digital transformation and is now gaining more value from AI. 

“Most remain earlier in the journey. They have core system modernisation still underway, early rollouts of CoPilot licences, small sets of AI use cases with mixed impact, and investor messaging that signals the arrival of the AI era. The gap between ambition and realised value remains wide,” it said.

One example cited in the report is voice AI developed by PolyAI. 

According to Baringa, these systems have cut cost per contact by 93% and increased customer satisfaction by 15 points. 

The technology allows policyholders to make claims, payments and account updates through natural language conversations.

The report also said insurers are using conversational intelligence tools to review 100% of customer conversations instead of relying on small samples. 

Data is another major issue for insurers. Baringa says accurate and timely data is needed for AI to work properly across operations. 

When customer data can move across systems, insurers can improve pricing, renewal timing, claims handling, complaint routes and workflow management.

The report said insurers could eventually use more advanced AI systems to manage parts of the claims process from end to end. 

For example, AI could identify liability signals at first notification of loss, start the right workflow, flag possible fraud and update the customer. 

Human staff would still handle judgement calls, oversight and exceptions.