Scout InsurTech Interview with Mike Steele

Mike Steele is Principal Enterprise Architect at Westfield Insurance, where he focuses on modernizing enterprise technology and advancing the use of AI to drive operational efficiency and stronger business outcomes. Mike was interviewed by Bobbie Shrivastav, Founder and CEO of Solvrays. 

Bobbie: How should leaders separate real AI opportunity from all the noise in the market today?

Mike: I start with a simple question: is there a clear line of sight to measurable business value? If the answer is no, it is probably noise.

There is a lot happening in the market right now, and it is easy to get pulled toward what is new or highly marketed. But if you cannot connect an idea to a tangible outcome, it is not going to hold up over time.

From there, I look at three things. First is impact. What are we actually trying to move? Cost, growth, risk, experience. If we cannot define success upfront, it is not worth chasing because we will not know if it worked.

Second is readiness. AI only works if the data and processes around it are ready. Do we have clean, accessible data? Do we have a process that actually benefits from intelligence or automation? Or are we just layering complexity onto something that already functions reasonably well?

Third is appropriateness. I always ask if AI is really needed. There are still a lot of cases where a simpler, more deterministic solution can deliver similar value with less cost and less risk. AI should either unlock something new, meaningfully improve an outcome, or set up future value. Otherwise, it is just added sophistication without a clear return.

At the end of the day, it is about staying grounded in outcomes. AI should create real advantage, not just check a box.

Bobbie: Where is the biggest opportunity to clean up the operational “messy middle” in insurance?

Mike: The messy middle is that space between intake and outcome where everything slows down. It is fragmented, manual, and full of exceptions, and it is where most of the operational friction in insurance actually lives.

Submission to bind is a perfect example. You have unstructured data coming in from everywhere, disconnected steps across teams and systems, and people constantly chasing missing or inconsistent information just to keep things moving forward.

There are three areas where AI can really help.

First is unstructured data. Insurance runs on documents. Emails, forms, reports, spreadsheets. Historically, we have struggled to reliably turn that into usable data. AI changes that. It can extract, interpret, and structure information in ways that make it actionable. Once that data becomes usable, everything downstream gets easier and faster.

Second is orchestration. A lot of the friction comes from small, repetitive tasks. Reviewing submissions, identifying gaps, sending follow-ups, resolving exceptions. Individually, they are simple. At scale, they create drag. AI can take on parts of that work and also recognize when something is incomplete or out of pattern, then trigger the next step. That reduces cycle time and operational friction.

Third is decision support. This is not about replacing underwriters. It is about helping them. AI can interpret context, identify patterns, and guide next best actions so people spend less time sorting through noise and more time making informed decisions.

The real opportunity is shifting from a reactive process to something that flows. Cleaner data, fewer interruptions, and a more connected path from submission to outcome.

Bobbie: Where should companies be most cautious when applying AI?

Mike: Anywhere you are making decisions, you need to slow down and be thoughtful.

With traditional approaches, whether it is rules engines or decision tables, you can explain exactly why a decision was made. You can audit it, trace it, and validate that it aligns with your standards.

With AI, especially probabilistic models, that becomes more complex. You do not always have the same level of transparency, and that introduces risk, particularly in areas like underwriting or claims where decisions carry real consequences.

That does not mean you avoid AI in those areas. It means you apply it carefully. You put guardrails around it. You validate outputs. You keep a human in the loop where appropriate, especially early on.

It also means making sure the AI is grounded in your business context, not just general data. If it is not aligned with how your organization operates, it can drift in ways that are hard to detect.

If you cannot explain the outcome or stand behind it, it should not be making decisions on its own.

Bobbie: What needs to be in place for a digital employee to work at scale?

Mike: It starts with trust. If people do not trust it, it does not scale. And trust is built on visibility, control, and consistency.

That means you cannot think about a digital employee as a standalone tool. You have to think about it as part of an ecosystem that supports how work actually gets done.

You need deep integration into core systems so it can take action, not just respond. A true digital employee is not answering questions. It is moving work forward.

You also need a strong data foundation. That includes governed data pipelines and the ability to unify structured and unstructured data. On top of that, you need a shared understanding of the business so it operates with the right context across different functions.

A robust knowledge layer is critical. The system has to be grounded in accurate, current information. Otherwise, you introduce risk very quickly.

You also need orchestration. These systems are operating across multi-step processes, often over extended periods of time. They need to maintain context and know when to hand work off to a human.

Governance is a major piece. Every action needs to be explainable and auditable. Without that, you will not get adoption across the organization.

Then there is the operational side. Monitoring, feedback loops, ownership. These systems start to look less like tools and more like a workforce that needs to be managed, measured, and improved.

And one area that is just starting to emerge is behavior. How should these digital employees communicate? When do they escalate? What is acceptable? As they become embedded in the business, we will need to define expectations for them just like we do for people.

At scale, it is not about the model. It is about everything around it.

Bobbie: How do you move fast with AI without creating long-term problems?

Mike: You have to create space to move fast, but keep your foundation steady.

First, separate experimentation from production. Give teams room to test, learn, and prove value in low-risk environments. That is where innovation happens.

Second, stay disciplined on your core architecture. Data, security, integration, governance. Those things should not be reinvented every time a new model or vendor shows up. They are what keep everything stable.

Third, build for flexibility. The landscape is moving quickly, so you need modularity. You should be able to plug capabilities in and take them out without reworking your entire environment.

And finally, put guardrails in place early. Clear policies around data usage, compliance, and human oversight allow teams to move quickly without creating unnecessary risk or slowing themselves down with approvals.

The goal is to capture near-term value without creating long-term complexity that you have to unwind later.

Bobbie: How is the push for simpler agent experiences shaping your approach to AI?

Mike: Agents do not care about your systems. They care about how hard it is to get business done.

What they feel is friction. Switching between systems, chasing information, interpreting rules, trying to piece together what they need to move forward. That is where things break down.

So the focus with AI should not be on features. It should be on removing friction across the workflow.

That means embedding AI directly into how work gets done. Pre-filling data from submissions, guiding appetite and eligibility, summarizing risk, surfacing what matters in the moment instead of making agents go find it.

It also means reducing cognitive load. Agents should not have to interpret everything themselves or hunt for answers. The system should guide them toward the next best action.

If we get that right, everything else follows. Faster quotes, smoother binds, better service, and stronger relationships.

Automation is not the goal. Making it easier to do business is the goal.

Scout InsurTech Thanks Our Partners