Marsh Risk is testing a new AI-powered risk strategy tool with a select group of clients as it expands its broader push to embed artificial intelligence into commercial insurance analytics and decision-making.
The broker confirmed that an internally developed platform known as “Project Leapfrog” is currently in alpha testing, Insurance Business has learned.
The tool, which has not yet been formally launched, is designed to provide real-time, directional guidance on corporate risk strategy using large language models, actuarial analytics and loss simulation technology.
During an interview at RISKWORLD 2026 in Philadelphia, Marsh Risk commercial director John Davies (pictured) described the project as an evolution of the company’s newly launched Risk Companion platform, an AI-enabled suite of analytics applications unveiled this week.
“We wanted to see whether clients had a need for this kind of application,” Davies said. “Rather than fully building something and releasing it only to find nobody uses it, we thought we’d give clients a taste of what it could look like.”
The Leapfrog tool is designed to sit within the broader Risk Companion ecosystem and help companies rapidly model the insurance and risk implications of strategic business decisions. Marsh formally launched the first components of Risk Companion – Renewal Companion for casualty and property lines, along with Captive Companion – at RISKWORLD.
In one example outlined by Davies, a client considering an acquisition could ask how adding millions of personally identifiable information records would affect its cyber exposure and insurance coverage. The Leapfrog tool would pull from existing client cyber models, assess current security controls, evaluate policy wording and rerun exposure scenarios in real time.
Risk Companion itself represents a significant overhaul of Marsh’s existing analytics infrastructure. The platform consolidates previously separate analytic engines for casualty, property, cyber and directors and officers liability into a unified “Risk Cortex” architecture powered by AI and Marsh’s extensive internal data sets. Davies said the company’s scale gives it an advantage because the models rely heavily on exposure, premium and claims information accumulated across its global client base.
The development reflects a wider shift across the insurance industry as brokers and carriers race to integrate generative AI into underwriting, risk analysis and client advisory functions. Marsh executives said the company’s goal is not to replace human expertise, but to accelerate analysis that previously required days of manual actuarial work.
Project Leapfrop would allow clients and brokers to test insurance program structures, adjust assumptions and compare outcomes in real time rather than relying on traditional back-and-forth actuarial modeling.
However, the deployment of AI in risk strategy introduces its own governance concerns. Even highly specialized AI systems remain vulnerable to hallucinations and inaccurate outputs, as well as embedded bias in historical claims data.
Davies acknowledged ongoing concerns about AI reliability but said Marsh is intentionally limiting certain use cases to tightly defined tasks. “It’s a guide, not a guarantee,” he said. “It still needs to be checked. It’s really about reducing manual workflow for the client.”
Client response to the new tools has been strong, according to Marsh executives, particularly among organizations already experimenting with generative AI technologies such as ChatGPT and Claude.
“We’ll press ahead and expand… based on the feedback we’ve had (at RISKWORLD),” Davies continued. “Building something in a vacuum without listening to clients doesn’t work. If we’ve got 20 or 30 clients volunteering to beta test it, that tells us there’s interest. Then we can shape it around what clients actually want instead of what we think they want.”
