In August 2024, just a day before flying to China, I sat in a closed-door session with the board of a respected regional insurance group. The CIO, flanked by contracted AI experts, unveiled a sleek demo: a chatbot for claims processing that cut average handling time by 50%. Applause filled the room. The directors were visibly pleased that innovation was alive and well, they thought.

As the board’s technical advisor, I interrupted the celebration with a simple question: “Who on this board understands how it works, what data it uses, or what risks it introduces?” No answers. Not one hand. Not even from the Audit Committee Chair. That was the moment I knew the board was impressed by the outcome but blind to the input. It was a classic case of being dazzled by performance while ignoring the black box that produced it.

Many boards today love to posture about innovation.

They add “AI” to the strategy document and mention “automation” in AGM speeches.

But the reality? Most are wholly unprepared for the governance burdens of strategic and ethical AI integration. In insurance, especially, AI now influences core functions underwriting, claims scoring, and fraud detection. Yet these models are often trained on biased, incomplete, or poorly governed data. And boards? They are asleep at the wheel.

The board’s role isn’t to code algorithms. It’s to ask the hard questions that pierce the veil of complexity. The real challenge isn’t approving AI, it’s supervising its integrity.

advanced divider

"Boards that don’t ask these questions aren’t delegating they’re abdicating. And in the realm of AI, abdication isn’t benign.” 

Tweet
advanced divider

What boards should be asking, every quarter:

  1. What decisions are being delegated to AI, and why?
  2. What’s the audit trail for those decisions?
  3. How is bias being tested, and by whom?
  4. What happens when the model fails?
  5. Who owns the ethical consequences?

Insurance boards are especially vulnerable. AI errors can mean wrongful claim denials, unjust premium hikes, or failure to flag fraud. One biased model can undo years of brand trust and trigger regulatory wrath.

To avoid this, boards must adopt a dual-lens oversight approach:

  1. Strategic: Does AI investment align with long-term resilience and customer trust?
  2. Ethical: Are models transparent, fair, and independently stress-tested?

Leadership Tool, The AI Oversight Radar

A five-question boardroom checklist, monthly:

  1. What models are currently used, and what decisions do they influence?
  2. What data governance policy controls the quality of inputs?
  3. Has any model outcome triggered regulatory scrutiny or client complaints?
  4. What level of assurance has been independently verified and by whom?
  5. Who serves on the AI Ethics and Risk Committee, and do they report directly to the board?

Boards that don’t ask these questions aren’t delegating, they’re abdicating. And in the realm of AI, abdication isn’t benign. It’s a time bomb with a silent countdown. Wake up, or be disrupted from the inside out.