An article by Elol Jacoby, co-founder and CPO at Notch,  adds more depth to the article I published yesterday of the dangers of digital agents proliferating across the insurance value chain. If a large claim has been rejected by digital agents and the claimant turns to litigation a judge will be waiting to hear from insurers 

This is one reason why Robert Pick,  Group Deputy CITO at Tokio Marine Group , when asked recently: “What is the one thing you need to make AI successful in your enterprise?”  answered with one thing and, actually, just one word: "Time."  ⏱️  

🕛 Time for AI safety and control to catch-up with AI capabilities
🕐 Time for our team to incubate deep technical and business understanding of AI
🕑 Time for AI providers to sort out token and credit madness and give us predictable cost structures
🕒 Time to establish how to properly measure value and outcomes, and relate them to AI expense
🕔 Time for our platform and system partners to mature their AI feature sets and offerings
🕕 Time for our excellent solution partners to flesh-out their AI practices
🕖 Time for our internal business partners to understand deeper in-transaction opportunities for AI acceleration
🕗 Time for org development and training colleagues to help us sort out the new skills and revised roles resulting from AI adoption
🕘 Time for auditors to sort out how to audit AI-based processes on a practical basis
🕙 Time for regulators and standards bodies to opine on all of this
🕚 Time for all of us to absorb these changes while still retaining control of all the things we do best (and our essential humanity!)
 

Over time POCs, trials, IT projects have often lead to what McKinsey termed ‘pilot purgatory’ where that 🕚 Time was not granted. Maybe FOMO leading the CEO and CRO to pressure for fast pilots without giving the CTO/CIO time to plan and resource. Annual reports, press releases, enterprise blogs want to show an insurer is ahead of the curve to leverage AIs but please give it time.

I was once parachuted into a global project that had stalled and was pleased to be given this advice by Chris Surdak who has leveraged AI across many industries from NASA to Web 4.0 . Ensure:-

1. Executive adoption decisions are based on something other than FOMO
2. Prove compelling use cases that will tackle real problems and/or leverage significant opportunities
3. Honest and complete business case ROI
4. Systems engineering thinking (Chris was involved in life & death decsions with space travel)
5. Effective test campaigns-measure outcomes rigorously
6. Ensure users buy into the vision and the experiments from early on
7. Understanding of the technology's true capabilities & the technology's true limitations
9. A backup plan if the experiment fails
10. Self-reflection over prior failures; what has changed in our approach this time?
11. Have the technology partners you chose ‘eaten their dog food’ and proved they can deliver the outcomes you desire using the tools they promote?

Combine Robert's and Chris's advice gained in the frontline of technology deployment and you will have a higher success ratio in operationalising and scaling any transformation project. You will then enable your company to answer the questions the judge and court asks with confidence.

What information was available and part of the decision at the time?

  • Which policy cover and other documents did the AI review?
  • Which policy language did it consider?
  • What did the AI recommend and what rules did it follow?
  • Was the recommendation accepted, rejected, or modified and on what basis?
  • Did a human set explicit rules and were these followed by digital agents?
  • Did an authorised human review the decision?
  • Did the human(s) have the intent to validate the decision or was it just box ticking?
  • What action was ultimately taken?
  • Why was that action appropriate?

Barry Rabkin and Jim Mitchell in “The Tool Master” detail why you must be able to asnwer those questions (see link in references at end of this article).

Jacoby states:-

"Many organizations use “human-in-the-loop” as shorthand for safe AI, but human review by itself is not a complete governance strategy. 

A human can approve a recommendation without fully understanding the evidence behind it. A human can miss a flawed output. A human can override a system without leaving enough context. A human can become a rubber stamp when volume increases.

The important question is not only whether a human was involved.

The important question is whether the institution can prove what happened, why it happened, who was responsible, what controls were applied, and whether the final decision was appropriate.

That requires more than review.

It requires traceability, policy enforcement, permission controls, escalation logic, version control, and audit-ready records that can be understood by business, legal, compliance, and regulatory stakeholders."

Rabkin and Mitchell say the same thing.

Follow the advice and make sure the judge does not make a decision against you that results in loss of good reputation, fines, and possibly accommodation in a US penitentiary or one of His Majesty's Prisons.

References

The Tool Master: An Insurance-Focused Agentic AI Thought Piece Barry Rabkin & Jim Mitchell

The AI Audit Report is no longer optional Elol Jacoby

When automous digital agents fight each other and forget an insurer's goals and intent M Daly