Written by Sabine VanderLinden 

Carriers report implementation. Brokers report hours. In 2026, that difference is worth 43 points of revenue growth, and clients have started switching over it.

Key Takeaways

In November 2023, Marsh McLennan gave 90,000 people an AI assistant at once.

It was called LenAI, built in Dublin with Oliver Wyman Digital on privately hosted models, and it went to the whole firm: not a lab, not a cohort, not a center of excellence with a steering group and a quarterly readout. Everyone. Today, it handles more than 700,000 queries a week.

The number that matters, though, is much smaller than that. Early adopters within Oliver Wyman report getting back about 8 hours a week.

Eight hours. A full working day, returned to one person, every week, for three years now. Multiply that by the people around them, and you arrive at the firm’s own estimate: roughly 1 million hours a year. But start with the eight, because that is where the advantage actually lives: not in the license, not in the architecture diagram, in somebody’s Tuesday.

Now ask what your own people got back this year.

What is a Frontier Operating Model? A Frontier Operating Model is the organizational design that lets a company run human-agent teams within live, revenue-producing workflows: legacy systems decoupled, one workflow at a time; venture partnerships that compress testing from years to weeks; and governance that clears at the speed of deployment rather than the speed of the committee calendar.

This piece examines the model, the broker-carrier AI adoption gap, venture-clienting methods such as DIVAAA and our venture-clienting framework, governance friction, rising client pressure for modern tools, and the practical steps to move quickly without getting trapped in implementation theater. It is what converts AI spend into operating advantage. It is also measured in hours returned, not tools deployed.

The gap has a number now, and it is 43 points

For two years this argument has been made with adjectives. It no longer has to be.

Grant Thornton’s 2026 AI Impact Survey, 950 business leaders across ten industries, fielded between 23 February and 18 March 2026, with an insurance subgroup of 100, found that organizations which have fully integrated AI into their workflows are nearly four times more likely to report revenue growth than those still piloting: 58% against 15%. Grant Thornton named the condition the “AI proof gap.”

43 percentage points. That is the distance between adoption and implementation, priced.

And the governance finding underneath it is the one that should be read twice: 44% of insurance executives said governance or compliance problems had contributed to an AI project failing or underperforming, and only 24% were very confident they could pass an independent AI governance review within 90 days. Three-quarters of the industry is running AI it could not currently prove it controls.

This is clearly an operating-model problem wearing a technology problem’s clothes.

Insurance carriers report implementation. Brokers report hours.

Look carefully at what each side is actually counting, because the two industries are answering different questions and reporting the answers as though they were the same.

Deloitte’s survey of 200 US insurance executives, fielded in June 2024, found 76% had implemented generative AI in at least one business function, rising to 91% among carriers above $10 billion in revenue and falling to 58% in the $100m to $500m band. Set that against Perspective AI’s May 2026 report: 64% of US agencies now run AI in at least one live workflow, up from around 38% in 2024, led by quoting at 71%, lead intake at 58%, claims handling at 49% and customer service at 44%.

I am not going to pretend those two numbers are a fair fight. The carrier figure is two years older, and anyone setting them side by side as a snapshot is selling something. What survives the date gap is the verb.

“Implemented in at least one business function” is a status. It is satisfied by a proof of concept, a governance framework, a signed license, a completed pilot, a press release. “Running in at least one live workflow” is a location. It says where the thing is: inside quoting, inside intake, inside the work that produces revenue on a Tuesday afternoon.

One shows up on the expense report. The other shows up in the 58%.

And the size data kills the last comfortable excuse. Adoption among agencies with more than 25 producers runs at 91%: the same figure Deloitte recorded for $10bn+ carriers two years earlier. Scale is not the constraint. Nobody with a large balance sheet is being held back by the balance sheet.

If you lead innovation inside a carrier, none of this is news to you: it is your week. Innovation theater is exhausting. Orphan pilots are demoralizing. You have watched good ventures die in procurement and presented the same use case to three successive committees, each of which asked for the deck to be shortened. The conviction was never missing. What is missing is an operating model that does not treat every single deployment as an exception requiring its own act of parliament.

What Frontier transformation looks like at the renewal desk

Here is the shape of it, at the scale where it is actually felt.

A regional brokerage put an AI workflow across roughly 3,200 renewing commercial accounts. Standard renewal triage had scored the book and flagged the risky ones. The workflow read the same book and surfaced 287 accounts at risk that the triage had scored green, accounts nobody was going to call, because the process said they were fine. Producers called them. 71% were retained. Retention on the book rose 9 points.

(Sourcing note: this case is drawn from 2026 broker AI-workflow research, and the firm is not named. Treat it as illustrative of the mechanism rather than as an audited result.)

Nothing in that story required a platform migration. It required one workflow, one decision to let a system read the book, and a willingness to act on an answer that contradicted the existing process. That last part is the whole Frontier Operating Model in a sentence: the model is only worth what you are willing to do when it disagrees with you.

Those 287 accounts existed inside every brokerage running the same triage. Most of them are still sitting there, still green.

The Frontier Firm route runs through the venture client model

Carriers do not need to become brokers. They need to borrow the broker’s operating logic, and there is a proven mechanism for doing exactly that without a transformation program.

The venture client model, taken to industrial scale by corporations such as BMW’s Startup Garage, Zurich Innovation Championship and ERGO ScaleHub, has one defining move: the corporate becomes a paying client of a young venture’s technology inside a real workflow, first and fast, before the strategic review, not after it. No equity, no committee, no 18-month integration. A purchase order and a problem.

Applied to insurance, that means decoupling one workflow from the legacy core and letting vetted ventures run AI microservices against it under production conditions, mirroring the venture-client and ecosystem plays highlighted at the Global Insurtech Summit on the industry’s digital future. This is the discipline our DIVAAA™ pathway enforces: validation gates that move a venture from pilot to production contract on evidence rather than enthusiasm, so pilots stop being theater and start being procurement.

The prize is the Frontier Firm profile, based on Microsoft’s 2025 Work Trend Index, measured across 31,000 workers in 31 countries: employees at Frontier Firms were 71% likely to say their company is thriving, compared with a 37% global average, and 55% likely to say they could take on more work, compared with 25% globally.

Human-agent teams handle triage, drafting, and reconciliation, allowing AI to take on routine tasks while humans provide empathy-led, high-value claims and service experiences. Underwriters and claims professionals concentrate judgment where it compounds, positioning their organizations to compete with agentic frontier firms powered by AI coworkers.

Firms structured this way test faster, learn faster, and compound faster. Firms that are not will discover that scale without speed is just weight, a pattern explored in depth in our analysis of why many AI strategies and startup collaborations stall.

The playbook: four moves this quarter

1. Pick one revenue-adjacent workflow and give it a 90-day production deadline. Quoting, submission triage, renewals. Not a sandbox. Production, with guardrails. Then hold the date as you would for a regulatory deadline.

2. Stand up a venture client lane. One page of terms, one accountable commercial owner, 30 days from first meeting to paid pilot. If procurement takes longer than the pilot, fix procurement first; that is the project.

3. Measure hours returned, not pilots launched. Ask your COO for one number every Friday: how many revenue workflows ran with agents in the loop this week, and how many hours came back. A number that arrives weekly changes behavior; a quarterly dashboard changes slides.

4. Book the governance review before you need it. Grant Thornton found only 24% of insurance executives are very confident they could pass an independent AI governance review inside 90 days. Find out which quarter you are in now, while the answer is still cheap and private.

What this costs, in the only currency clients use

The uncomfortable part of the Zywave finding is not the statistic. It is where the statistic gets delivered.

More than 1,400 US employers told Zywave in July 2026 what would make them change brokers. The top three are what they have always been: slow response times, inconsistent communication, and a lack of strategic advice. But for the first time, “failure to leverage modern technology and AI-powered tools” joined them on the list. As Zywave CEO Martin Simoncic put it: “AI is no longer optional infrastructure for brokers. It’s becoming a visible part of how clients judge value.”

Read that as an operating fact rather than a survey result. It means the gap is no longer discussed in your strategy offsite. It is discussed in your client’s, without you in the room, in the sentence that begins “they still take four days to turn a quote around, and I don’t think they’ve moved on any of this.”

You will not hear that sentence. You will hear its consequence, at renewal, from an account you were confident about: one of the green ones.

That is the loss sitting behind the committee cycle. Not a market-share chart. A person you have known for nine years explaining, kindly, that they have moved.

Meanwhile somebody’s producer is walking into that meeting having got their eight hours back, and using every one of them on the account.

The Frontier Operating Model is not a destination the industry arrives at together. It is being built right now, quarter by quarter, by whoever is willing to put one workflow into production before the committee finishes deliberating. Brokers started early because the market gave them no choice. Carriers still have a choice, and choices expire.

The real question isn’t whether carriers can move at broker speed. It’s which carrier proves it first, and whether it’s yours.

Where does your organization actually stand on the path to a Frontier Operating Model? Connect with us to share your views at [email protected].

Frequently asked questions

References

1. Marsh McLennan. "Marsh McLennan Develops New Generative AI Tool." November 2023
2. Oliver Wyman. "LenAI: A Generative AI Tool By Marsh McLennan." September 2023
3. Life Insurance International. "Marsh McLennan launches new generative AI tool for employees"
4. Consulting.us. "Marsh McLennan launches AI tool LenAI" 
5. CIO Dive. "Marsh McLennan gears up for massive AWS migration" 
6. Predibase. "How Marsh McLennan saved over 1 million hours of team time with agentic AI" 
7. Grant Thornton. "Insurance insights: 2026 AI Impact Survey Report." 2026
8. Grant Thornton. "A widening 'AI proof gap' is emerging, but well-governed AI is showing results." Press release, April 2026
9. Grant Thornton. "2026 AI Impact Survey Report." 2026
10. Businesswire. "Grant Thornton survey: A widening 'AI proof gap' is emerging, but well-governed AI is showing results." 13 April 2026
11. Insurance Journal. "Grant Thornton: Insurers See AI Gains but Face Governance Gap." 30 April 2026
12. Insurance Business. "Widening 'AI proof gap' exposes weak governance behind board-level enthusiasm"
13. Deloitte Insights. "Scaling gen AI in insurance." Survey of 200 US insurance executives, fielded June 2024
14. Perspective AI. "AI for Insurance Agents in 2026: Adoption Hit 64%. The Industry Data Report." May 2026
15. Perspective AI. "AI for Insurance Agencies in 2026: From Lead Capture to Renewals." 2026
16. Layer3 Labs. "AI Workflow Automation for Insurance Brokers (2026)"
17. BMW Startup Garage. Venture client model
18. Microsoft. "The 2025 Annual Work Trend Index: The Frontier Firm is born." Official Microsoft Blog, 23 April 2025
19. Microsoft WorkLab. "2025: The year the Frontier Firm is born." April 2025
20. Microsoft. "Executive Summary: Work Trend Index Annual Report." April 2025
21. Zywave. "AI Emerges as a Defining Force in the Broker-Client Relationship, Zywave 2026 Broker Services Survey Finds." July 2026
22. Insurance Business. "Clients now expect brokers to lead on AI, not just advice"
23. Digital Insurance. "Employers would switch brokers because of AI, other tech: Zywave" 
24. Insurance Journal Research. "2026 Broker Services Survey Results"
25. Insurance Edge. "Zywave Releases 2026 Broker Services Survey." 28 July 2026
26. Microsoft. Jim DeMarco, author archive, Microsoft Industry Blog