Modernizing Insurance Pricing From Excel to Explainable AI - Thomas Holmes of Akur8

Holmes identifies the dual challenge of implementation and governance as the primary friction points preventing modernization. Excel's perceived benefits (transparency, flexibility, ease of learning) are actually misconceptions that create resistance to change. The key barrier is that new tools must be unequivocally better while maintaining actuarial rigor and explainability.

• •Excel is surprisingly opaque and inflexible despite being perceived as transparent and easy to use
• •Actuarial teams will hard-stop any solution that doesn't provide actuarially sound results
• •New methodologies often fail the governance test due to complexity and lack of explainability
• •The status quo persists due to decades of training momentum, not actual superiority of tools

Generic, off-the-shelf AI algorithms fail when applied to insurance problems due to the nuanced nature of actuarial work where data often 'lies' and requires expert interpretation. The scattered approach of layering AI on top of existing Excel workflows or providing generic copilots creates adoption issues rather than solving fundamental problems.

• •Actuarial work requires identifying when data is misleading - a key reason actuaries have jobs
• •Off-the-shelf AI lacks the insurance-specific nuance needed for real-world considerations
• •Avoid 'stilt AI' that builds on top of flawed existing systems rather than replacing them
• •Question whether legacy tools like Excel needed to exist in the first place before modernizing

Successful AI modernization requires opinionated frameworks with built-in guardrails rather than unleashing flexible, complex AI tools without oversight. The vision for modernization must balance AI's power with structured paths that prevent governance nightmares and maintain actuarial control.

• •AI's flexibility, complexity, and power create a governance nightmare without proper frameworks
• •Implement AI within defined paths and guardrails rather than turning it completely loose
• •Build opinionated use cases for AI rather than generic applications
• •Prevent mistakes and lack of oversight by constraining how AI is used within processes

Holmes advises starting with clear problem statements rather than fear-of-missing-out driven AI adoption. Begin with incremental wins, choose the right form of AI for specific problems (bespoke vs. generic), and focus on getting the data foundation correct before scaling.

• •Reject FOMO-driven implementations - start with specific problems that need solving
• •Get small wins first to validate you've chosen the right AI approach for your use case
• •Match AI type to problem: LLMs for some tasks, visualizations for others (e.g., rate change analysis)
• •Fix data foundation first - AI is an input-output machine that requires quality data

Holmes advocates for 'reinventing the wheel' - taking proven processes and modernizing them with better tools rather than completely ripping out everything. The goal is to remove rating calculations from Excel and put them into systems with better governance, comparison capabilities, and controls while keeping what works well.

• •Evolution example: wheels progressed from stone to rubber to Mars rover tires - continuous improvement
• •Don't rip out all existing processes - reimagine how to do them better with new capabilities
• •Moving from Excel to modern systems unlocks better tools for comparison and governance
• •Keep processes that work well while innovating on the implementation layer

Successful implementation requires addressing distinct concerns across three key groups: actuarial teams need actuarial soundness, leadership needs quantifiable benefits beyond 'cool factor,' and IT needs scalability and integration plans. Each group has deal-breaker requirements that must be met.

• •Actuarial teams: actuarial soundness is non-negotiable regardless of speed or test statistics
• •Leadership teams: need cold hard numbers and specific benefits - 'cool factor' doesn't work
• •IT teams: require answers on scalability, upkeep, and integration with current processes
• •Listen to all concerns with an open mind - each function has legitimate requirements

Leadership buy-in varies dramatically between insurtechs (flexible, innovation-focused) and legacy insurers (established processes, enormous inertia). Legacy organizations often publicly support innovation while privately maintaining status quo until pain points become severe enough to necessitate change.

• •Insurtechs have advantage: no legacy tech stack, shorter premium history, built-in innovation culture
• •Legacy insurers face inertia from established processes that are working well
• •Quantifying benefits is harder when replacing current processes vs. building new ones
• •Many teams are 'two-faced': openly supporting innovation while actions maintain status quo until pain is sufficient