Insurance organizations are racing to expand AI – 86 percent plan to raise their AI spending this year, with generative and agentic tools topping the list, according to Accenture research cited by Insurance Business – yet few can match the scale Marsh has already reached. Puneet Satyawadi (pictured), chief operating officer of Marsh, the global brokerage owned by New York-based Marsh McLennan, describes an adoption curve that now shapes where the company invests next.
Marsh McLennan's Dublin Innovation Center built LenAI, the firm's proprietary generative AI assistant, together with its Oliver Wyman consulting arm, and it runs a private, internally hosted version of OpenAI's models. The firm made a deliberate choice to put it in everyone's hands. “Our approach was to democratize the availability of infrastructural AI tools,” Satyawadi said; Marsh made LenAI available to every colleague across more than 80 countries, and he says it integrated seamlessly across all regions.
Early hesitation gave way once training and a completion badge drove awareness, and usage has since climbed sharply. “I used to get happy in the early days when every month we had 3,000, 5,000 prompts,” said Satyawadi. “Last week we had a million.” The company has since rolled out LenWork, a newer platform, to 27,000 people within weeks, and colleagues have built roughly 2,000 reusable capabilities for similar job profiles inside LenAI – up from about 1,800 a month earlier – which an internal innovation group then turns into funded investments.
That behavioral data has become a planning tool. “We now know where our colleagues are using AI,” Satyawadi said, calling it “tremendous information” for shaping “some of our large chunkier investments to capture those themes.” Rather than guess where to place its bigger bets, Marsh reads them off actual usage. Client-facing teams lean on the tools to sharpen coverage advice, drawing on the broker's data stores to tailor recommendations. “Being the largest broker in the world gives us more data than anyone else,” Satyawadi said, and putting the power to analyze that data in the hands of client-facing colleagues is “a massive differentiator.”
Marsh has scaled that idea into what it calls a coverage intelligence platform, and more than 3,000 colleagues now query a structured repository of claims data to judge how carriers perform and to steer placement.
Asked where that data heads next, Satyawadi sees two uses. The first is better insights for clients as Marsh advises them. The second is training the firm's own agents: “We're seeing examples of agentic deployment,” he said, run against a defined risk appetite with humans kept in the loop, and he expects that appetite to widen as the technology proves out.
Those tools sit inside a broader push Marsh laid out on its second-quarter 2026 earnings call, where the firm tied its AI work to growth, productivity, and efficiency. Rival platforms chase the same middle market, as reporting on a new brokerage platform targeting the mid-market growth gap shows.
That activity invites a comparison with Marsh's peers: Aon publicly reports large absolute information-technology spending – more than $500 million a year since 2022 – but only modest year-over-year growth, a slim share of total operating expenses; Arthur J. Gallagher meanwhile has raised its technology operating expenses steadily, including a sharp pandemic-era pivot. Marsh McLennan, by contrast, publicly reports restructuring-specific technology charges that rise and fall year to year. Satyawadi rejects that the brokers absolute technology spending is episodic: “[Our] strategy and approach does not allow for incremental or sporadic thinking,” he said.
Scale like that is rare and expensive. BCG research puts enterprise-scale AI deployment at $50 million to $100 million a year and finds only about 10 percent of firms reach scaled deployment in any single function.
On the fence between internal or external technology development or procurement, Satyawadi said Marsh tilts firmly inward; The firm leans hard toward internal builds. “We would like to have our future firmly in our hands,” Satyawadi said, prizing optionality between vendors. “We don't want to be committed to a single LLM or a single LLM provider,” he said, even as LenAI today runs on a privately hosted instance of OpenAI's models.
When Marsh does turn to vendors, Satyawadi said the firm is very clear about the broker value stream, and he urged vendors before they pitched to understand that chain: onboarding clients, placing business, servicing policies, handling claims, and carrying fiduciary accountabilities.
That clarity does not shut vendors out. Marsh keeps a vendor network that drives ideation and solves pieces of its strategy, and Satyawadi insists the source of a good idea does not matter: “Size doesn't matter, the quality of the idea matters.” For instance, Marsh has iterated with a two-person team on one idea, and with a global enterprise on another (document ingestion, in that case).
The broker is also pushing into agentic AI, where software agents execute steps of the workflow. “If I take a look at the number of transactions that are flowing through agents at the moment, they will be in the hundreds of thousands,” Satyawadi said, adding that “dedicated human-in-the-loop teams” govern them and that Marsh's diligence there “is quite robust at the moment.” He expects that balance to shift as the firm's risk appetite matures: “Our appetite would evolve, and I think the nature of the human-in-the-loop team we have interacting with these agents is going to change” – an evolution that sharpens the liability questions now surrounding autonomous AI agents.
Recent deals show the buy side of that balance. In July 2026, SS&C Technologies said Marsh would use its SS&C Blue Prism WorkHQ platform to scale agentic automation, building on 130 digital agents already in place. A year earlier, Marsh McLennan named Amazon Web Services (AWS) its preferred cloud provider and migrated the bulk of its infrastructure, work that also strengthened LenAI. Paul Beswick, chief information and operations officer at Marsh, has cast those partnerships as a way to move automation beyond basic scripting into governed agentic workflows across historically siloed data. The spending fits a wider surge, one Insurance Business charted when it asked whether AI's acceleration has left insurers ready.
For all the platforms and vendors, Satyawadi returns to a point he wants buyers to hear: Marsh does not treat AI as a technology problem. “It has a very significant people component,” he said. Every deployment, in his telling, carries a change-management plan alongside the technology, because adoption – not the model – decides whether the investment pays. On the evidence of a million prompts a week, Marsh's colleagues have made that call for themselves.
