The Shift in the Conversation
A year ago, I was telling insurance leaders to start pragmatically. Pick one problem, run a pilot, prove value, then scale what works.
That advice still holds for foundational learning. But something shifted in the past twelve months that makes it insufficient.
The question is no longer "where should we pilot AI?" It is "are we ready to redesign how we work?"
Agentic AI, machine learning, and integrated analytics are not additive to your existing operating model. They are incompatible with it. The companies that try to bolt these capabilities onto 2015-era process design are discovering that pilots work fine, but scaling them does not. The silos push back. The governance breaks. The measurement frameworks become meaningless.
I am now having very different conversations with leaders. The question is no longer "where should we pilot AI?" It is "are we ready to redesign how we work?"
The Pilot Paradox
Here is what I keep seeing: A team runs a successful pilot. Application intake automation reduces cycle time by 30 percent. Customer satisfaction improves. Cost per transaction drops.
Then the team tries to apply the same logic to claims triage or underwriting support. They expect 30 percent improvement. They get 8 percent. Or the initiative stalls entirely.
Why? Because those pilots were optimizations of existing process. They made current workflows faster. But they did not change the underlying thinking.
When you introduce agentic systems, machine learning, and real-time behavioral data into the mix, you cannot just accelerate old decisions. You have to redesign what decisions actually need to be made, where, and based on what signal.
That redesign cannot happen inside a pilot. It requires rethinking the entire decision architecture.
What Agentic AI and ML Actually Demand
Let me be direct about what we are talking about here.
Agentic AI means autonomous systems making decisions and coordinating across workflows. That works only if you have clarity on what decisions should be made by agents versus humans, when escalation happens, and what data feeds those decisions. Most insurance companies do not have that clarity. Decisions are distributed, rule-based, and opaque.
Machine learning means building predictive models on real customer and risk data. That only creates value if you have data governance that actually works, if your data is integrated across silos, and if your process is designed to use the signal the models produce. Most insurance companies have fragmented data, inconsistent definitions, and no mechanism to act on predictive insight at scale.
Integrated analytics means seeing patterns across the entire customer lifecycle, not just individual transactions. That requires decision redesign. If underwriting, pricing, servicing, and claims are still making independent decisions based on isolated data, analytics will not change anything.
These are not technology problems. They are organizational and strategic problems.
The Real Work
I have been in several rooms now where a CIO and COO are debating AI budgets while their underwriting team and claims team are still in separate buildings running separate processes. The CIO wants to build an AI platform. The COO wants to cut cost. Neither is asking the question that actually matters: What would our decision-making look like if we designed it for clarity, speed, and risk insight instead of designing it for ease of hand-off?
That question is uncomfortable. It implies change. It implies rethinking how work is organized, who makes decisions, and what accountability looks like.
But that is the work. Not the technology. Not the pilots. The organizational redesign.
The companies winning here have done that work first. They have mapped their decision architecture. They have asked what should be automated, what should be augmented with ML, what requires human judgment, and where the hand-offs should be. Then they build the technology into that new model.
Everyone else is still treating AI as a tool to make existing work faster.
What This Changes for Leaders
If you are going to go beyond pilots, you need to get clear on three things.
- Your decision architecture. Not your process architecture. Not your organizational chart. Your decision architecture. Where are consequential decisions made? What information should inform them? What signal do you currently lack? How do those decisions connect across the customer lifecycle?
- Your data strategy. Not your data platform. Your data strategy. What data actually matters for better decisions? Do you have it? Is it integrated? Are there governance or privacy obstacles? Do you know what your model quality baseline is?
- Your operating model. Not your technology operating model. Your business operating model. If you redesigned how work flows based on clarity of decisions and availability of signal, how would it differ from today? What would you need to change organizationally to make that new model work?
Answer those three questions seriously. Then build AI into what you discover.
The Market is Not Waiting
I keep running into insurance leaders who are watching the market move and feeling like they do not have time for redesign. They want to move fast. They want incremental wins.
That instinct is understandable. It is also becoming a liability.
The companies that are redesigning their operating model around AI and data are moving faster on the things that actually matter. They are making better underwriting decisions. They are retaining customers better. They are spotting fraud and mispricing earlier.
The companies that are still optimizing old workflows are getting marginal improvements that fade over time.
The Question Beneath the Question
Here is what I think leadership teams are really wrestling with: If we redesign this, are we admitting that how we have been doing it was wrong?
The answer is yes. In some cases, it was. In others, it was right for its time and is now outdated.
That is not a failure. That is how organizations evolve. But it requires humility and clarity about what has changed and why the old model no longer works.
The companies that move well here are the ones led by people who can say, "We built something that worked. Now the market and the technology have changed. We are redesigning."
Everyone else is still defending yesterday's choices while watching tomorrow pass by.
Questions Worth Sitting With
What would your decision-making look like if it were designed for speed, clarity, and signal instead of for organizational convenience?
Where are you currently optimizing for the wrong thing? Cost per transaction instead of customer and risk clarity? Speed instead of quality? Process efficiency instead of business outcomes?
What capabilities do you need to hire for that are different from what you have been hiring?
And most importantly: Are you ready to say that how you have organized work needs to change?