Credit: Alexandr Aleshin/Adobe Stock
Every few weeks, the same argument resurfaces: Is AI a bubble? Too much hype. Too much capital. Too long of a tail to make a gigantic investment worth it. Too many vendors "AI-washing" yesterday's automation. Maybe.
But if you work in protection services and risk management, that debate is a distraction. Even if the market cools or pulls back, the more important shift is already happening inside operations: AI is changing the unit economics of work. Not "someday." Right now. The real question isn't whether AI is overvalued. The real question is: What happens to underwriting, claims, investigations, service, and distribution when reading, writing, summarizing, and routing become near-zero-cost capabilities?
At its core, the protection and advisory sector is an industry of language and decisions. Workflows are complex, involving people interpreting and moving information, then deciding the next steps. This process is interwoven with a multitude of potential artifacts, including policies, endorsements, submissions, loss runs, adjuster notes, medical records, email threads, PDFs, photos, videos, recorded statements, litigation documents, and regulatory forms.
All of these are used as people declare, capture, and infer data to facilitate the decision-making process. Current AI capabilities represent a tremendous opportunity to compress that cycle and leave as little stones unturned as possible.
The opportunistic shift: From "tools" to a new operating model
Most carriers and agencies are still treating AI like a feature: bolt it onto a workflow, automate a step, move on. That's why so many pilots stall out. You get a demo, a dashboard, and a handful of "time saved" anecdotes but then reality hits: exceptions, compliance, edge cases, and the messy human parts of claims and underwriting. That's where human consciousness needs to be injected back to preserve trust and accountability.
The deeper shift happens when AI turns workflows into systems of orchestration. Instead of humans doing the first draft of everything, AI does the first pass that can rely on deep research and strong knowledge foundations. Humans can then become editors, escalators, and fully fulfill their role as informed accountable decision-makers.
In practice, this could mean that submissions get triaged and summarized before an underwriter even touches them, claims intake gets structured instantly instead of living in scattered notes, coverage questions get answered faster because institutional knowledge is searchable in plain English or that routine customer service becomes a managed flow, not a queue of repetitive conversations.
Those outcomes are not the manifestation of a "bubble" but perspectives for a permanent change in how work gets done.
Why this matters specifically in P&C
Insurance has two non-negotiables: speed and trust. Speed because the best operators win on cycle time (quote-to-bind, FNOL-to-resolution, litigation containment). Trust because a single bad action or decision such as wrong denial, wrong payment, wrong communication creates regulatory, reputational, and legal blast radius.
AI increases speed dramatically and it also increases the consequences of sloppy execution. So until large language models have matured into tools that are flawless and self-sufficient, the winners won't be the companies that "adopt AI" the loudest. They'll be the ones that make measured and responsible use of what it can accomplish today and redesign their workflows with intent and guardrails and treat AI like a high-leverage junior teammate in need of tight supervision, not a magical oracle.
Three practical moves that help cutting through the noise
1. Pick work that is mostly language, not judgment: Start where the risk is low and the volume is high.
- Summarizing claim notes, recorded statements, medical records, and demand letters to streamline the documentation flowing through processes end-to-end.
- Drafting customer communications for review to preserve branding, voice and tone across all cases.
- Extracting key fields from submissions and endorsements and identify truly actionable elements at key moments throughout workflows.
- Building strategies and plans with recommendations in the interest of the "next best action" without auto-executing them.
- Implement a first line of service and communication with the customer to answer their questions, help structure and document their claims, follow through workflows and keep communications going at the pleasure of the customers.
Currently, adoption is not about replacing adjusters or underwriters. It's about automating the busy work and reducing the time they spend doing clerical interpretation so they can do actual value-added tasks involving judgment.
2. Define your "human-in-the-loop" rules like you mean it: This is where most programs fail because leaders want the benefits of automation but don't specify the boundaries.
- What can AI draft, what can it recommend, and what can it do?
- What requires a licensed human sign-off?
- What exceptions trigger escalation?
- What must be logged for auditability?
If true accountable leaders of AI initiatives do not define these rules, the organization will "vibe" its way into inconsistency and inconsistency is where regulators, plaintiffs, and customers live.
3. Treat AI-driven fraud as the next operational tax: The same tech that helps you automate will help bad actors scale fraud: synthetic identities, deepfake voices, manipulated photos, spoofed payment instructions. This isn't theoretical. It's a control problem.
- Verification steps for payment changes
- Multi-factor confirmation for high-severity claims actions
- Fraud models + AI-assisted human review workflows that assume "better fakes" are coming
- Staff training for secure adoption of the AI Policies is unavoidable so frontline teams understand the risk and are capable to recognize new patterns
AI delivered with such attention to change management will not just improve a team or a company's efficiency; it will raise the baseline of the threat landscape. If you're leading an agency, carrier team, or claims organization, there is an uncomfortable truth to be dealt with: the AI bubble debate is mostly about money. Industries should be talking about operating reality. If it doesn't happen with them in the driver seat, it will happen to them.
The organizations that win won't be the ones arguing whether AI is overhyped. They'll be the ones quietly rebuilding workflows around a new baseline using AI for the first pass, humans for accountable decisions, implementing controls for trust and metrics tied to cycle time, leakage, severity and customer experience. If there's a bubble, it's in the storytelling. It is fueled by insatiable demand for inference computing. The shift in work? That part is real and it's not going away.
Nicolas Genest is the founder and CEO of CodeBoxx.
Opinions shared in this piece are the author's own.
© Arc, All Rights Reserved. Request academic re-use from www.copyright.com. All other uses, submit a request to TMSalesOperations@arc-network.com. For more information visit Asset & Logo Licensing.