AI solutions chasing insurance problems – Insurtech musings 2024

Insurtech LV in 2024 was a superb platform showcasing the progress of technology across the Insurance sector. The focus on Ai and automation was not a surprise but the variance of opinions and degree of knowledge certainly was. I spent most of my time getting to the vision and value of AI in Insurance from executives at the show on AI and their plans was enriching and eye opening. The hunger to process data, detect patterns and make better decisions has never been more, however the desire to leverage the benefits that AI brings is unbridled mixed with a cautious realization that it cannot be rushed.

The best summary was from a mid west CEO who said “AI right now is a lot of solutions chasing a problem to solve” that really put the whole thing in perspective. Suffices to say neither the multitude of vendors at Insurtech nor the potential enterprises really have the complete picture or understand the complexities and challenges end to end, partially thanks to social media and part due to different vendors trying to position their part of the solution as the magical tool needed to harness the enormous potential of AI.

80% of the executives think AI challenges are data and siloed legacy systems – that is exactly the problem most don’t even realize the actual challenge of AI.

To repeat a highly abused cliché - enterprise AI is a journey and not an implementation – it is a marathon not a sprint. After 22 miles in a marathon if you realize you are in the wrong direction it is time to shut down and so would your AI journey be if you are not clear about the vision, checkpoints to reiterate and recalibrate the benefit realizations at mid points leading to a typical introspection of what went wrong. Computing the ROI is a very key first step – detailing the business case and RPO may not be a 100% accurate exercise at earlier stages but starting without a vision to what you are looking to achieve defined at a metrics / $ level is not good–we always over estimate what we will achieve in one year and under estimate what we can in five years.

The significance given to the business context is lower than due across all enterprises. Interesting use cases in underwriting, agency management , risk management , loss predictability and collaborative product design are being widely discussed but what is missed is that these demand completely different business contexts and approach. As much as business context is the clarity of the goal is it efficiency, cost optimization or speed. Sometimes these can be conflicting and prioritization has to be built in.

A blueprint that will define the problem statement well, identify the source of truth that can be leveraged, pick the best tools and partner are all critical to get an insight into whether this is a business problem solvable by AI or not. A good legacy repository of data or data warehouse does not guarantee an AI engine nor business success.

Every other integration and data analytics consultant is out there selling on data readiness as the magic potion for AI and there is some truth in it. Data readiness is really like saying you need a good piece of land to build a house – true without that there is no way a house can be constructed but at the same time if you spend millions on acquiring the land without a clear vision of the house and the building complexities you will end up nowhere. And spending millions in getting all your data AI ready is also not a good strategy – identify the data that will solve the problem first. Analyze if your historical data alone suffices or you need to expand the horizon and is the feasible?

The choice of model should always always be aligned to the business problem. Unfortunately in the Insurance industry today different models are at varying degrees of maturity and that shows why we should stress on the significance of the business problem first – almost every step aligns with it. Do you look at Generative AI or Econometric modeling or LLM or Learning engines – these are all designed and work well for different business scenarios. And this is a decision for the business executive team and not your partners, simply because the key insight required is the final desired outcome.

The elephant in the room is talent and people – everyone is under-estimating this but we have no doubt the talent strategy will be the deciding factor between success and failure. A good committed partner who values compliance and security as well as provides the support when things go off track is more important than just capabilities. Unpopular but common opinion among attendees was that North America has a plethora of partners building AI solutions for BFSI but we have not seen the same degree of maturity in partners who are equally adept at data management, business consulting and data science. The scenario lends itself to a strong AI outcome office for the enterprise that manages the multiple specialist vendors and herds them to the goal.

The partnership strategy should not ignore nurturing internal talent and helping the current workforce understand the AI vision and therein pushing for reskilling / training to make them AI ready. Also critical to separate the vendor / partner skills to inhouse skills to enforce knowledge bridges. Once the platform is set there will be multiple layers of training , retraining of the models, monitoring all better invested through internal resources or hybrid teams than exclusive partner dependency.

In summary there has never been an enterprise initiative that begets collective accountability and a collaborative effort between internal and external stakeholders as AI in Insurance. Easy to be overwhelmed by the enormity of the challenge and lose focus on the business goal.