As AI and ML advance, getting it right means perfecting the holy trinity: maximising business value, minimising operational cost and risk, and ensuring simple adoption.
Unfortunately, only some use cases can be optimised this way, often requiring a big compromise on one of the three.
This is a challenge I've faced while helping customers on their AI and ML adoption journey, with successes being few and far between, but those who succeed do so in a big way.
Machine learning in insurance is nothing new. We've seen its use in actuarial pricing for modelling risk, actuarial reserving and claims, and in predicting claims amounts more effectively.
Many have now started to dabble into Generative AI in insurance with various successes, in my previous blog I highlighted a number of these with document processing showing the most success.
The combination of unstructured data in insurance and the advances in generative AI and LLMs has opened up significant untapped areas for value creation.
One use case that excites me, which we've been working on with our customers, meets all three points of the holy trinity. It promises big successes by addressing a prevalent issue for many North American insurers, with a Workers' Compensation line of business.
Workers' Compensation insurance, similar to Employers' Liability Insurance in the UK, is mandatory for most businesses with employees.
It protects employers from financial liabilities arising from employee claims related to work-related injuries or illnesses, and ensures employees receive appropriate compensation and support during recovery.
A key area where this happens is injury creep. Injury creep, also known as "claim expansion" or "diagnosis creep," refers to the gradual addition of new medical conditions or body parts to an existing Workers' Compensation claim.
This occurs when an employee's initial work-related injury or illness is expanded to include additional diagnoses, treatments, or impairments that were not originally part of the claim.
In most cases, this is due to a natural progression of the injury and the claimant is entitled to their compensation. But in some cases, these additional injuries are not related at all and thus shouldnt be compensable.
Injury creep can significantly increase the cost and duration of a claim, as well as complicate the claims management process for insurers.
Claims with injury creep will naturally involve higher medical expenses, administrative costs due to increased oversight, and potential litigation expenses. With total incurred losses for state carriers around $18bn, this is no small feat.
We've been helping customers leverage generative AI to reduce these losses and expenses, focussing on two key use cases:
This has helped our customers:
These tangible successes have led our customers to explore other areas where this capability can help, such as:
This all sounds great. But how do we do this safely, and securely with good AI governance in place to ensure we have regulatory compliance? How do we ensure an accurate enough model provides trustworthy and reliable results that provide value and not making decisions that negatively impact human lives?
We’ve got the battle scars from encountering these challenges and devising solutions to overcome them, so you don't have to.
In the next blog we talk about approaches for AI governance frameworks – from tracking and storing user inputs, injecting safety prompts into user input to ensure responses align with organisational policies, to scanning and monitoring model outputs for bias and fairness – that ensure we can wrap this up safely (as well as keep the risk, compliance and governance teams happy).
.png)
.png)
