The state of AI in EB brokerages pt. 1

In-Woo Park

For the last year, most brokerage AI conversations have been stuck at the hype demo layer: a better summary, a faster email, a chatbot that can answer a question from a policy document.

That's changing with serious moves starting to show up in the core work of brokerages: how teams grow business, service clients, and increase service quality without adding headcount proportionally.

Some recent signals are worth taking seriously:

Brown & Brown + Anthropic, McKinsey, Accenture

In July, Brown & Brown announced an AI-first strategy with Anthropic, McKinsey, and Accenture. The plan is to bring Claude to roughly 23,000 team members, make Claude Code available across its engineering organization, and use a Value Management Office to track adoption, ROI, and controls.

What's interesting is how they've publicly pointed to a target of automating 25% of submission workflows without replacing teammates. It has also been direct that the larger benefits may arrive in years three, four, and five rather than the first quarter after a rollout.

That is a more mature framing than “we bought an AI tool.” It's more intentionally treating AI as a long-term capability: new infrastructure, workflow redesign, and a period where the organization has to learn how to use it.

Alight - $90M in projected annual savings

Alight is a large benefits-administration company. Though not necessarily a brokerage, its latest SEC filing is quite interesting.

Its new productivity program uses AI as a lever in a larger cost-and-capacity restructuring plan: automate more work, rethink outside-service spending, and reduce unnecessary overhead. Alight expects up to $90M in annual savings after the program is finished.

What makes this interesting is the framing: this is not another software license added to the P&L. Alight is treating AI as a restructuring capacity decision -- whether it can take on more work without adding the same amount of headcount and overhead.

Marsh McLennan - increasing speed of service through AI

Marsh’s AI push is less about doing the same work with fewer people, as productivity is only part of the AI pitch. CEO John Doyle has framed the opportunity around three things: productivity, efficiency, and growth.

According to a report on Marsh’s April earnings call, several thousand colleagues are using Claims IQ to analyze almost $200B of loss information for client advice, decision-making, and advocacy. The company’s document-ingestion tools are handling thousands of documents a week and have improved efficiency in those processes by 20%, Doyle said.

Put simply, Marsh is using AI to help its teams give clients better answers through faster access to the right loss information.

From our own pilots at Fidaris

From our own pilots at Fidaris, we are building a new type of capacity for brokerages, which we call the M+1 model. Our thesis at Fidaris is that in producer-led markets -- including EB brokering -- technology will increase the output, quality, and revenue per employee, so the focus is more around business growth instead of cost-saving.

This is achieved through our AI system that has access to unlimited memory across and agency's book, and the ability to execute work inside carrier portals, AMSs, CRMs, email inboxes, shared drives, benefits platforms, etc. where the work happens.

We're helping brokerages spend at least 40% less time per client. That new capacity gets re-allocated towards more clients, more focus on strategy, and more time towards growing the business, where we also give teams an unfair advantage.

Five questions worth asking before you buy anything

That said, there is still plenty of noise. Many folks struggle understand how or what to adopt in the first place.

If you're a smaller brokerage or a large national still considering running another AI pilot before deciding on a larger-scale rollout, here are 5 questions to ask:

If everyone in the team had a smart intern, what work would we offload to them? This is the first question we at Fidaris ask the teams we work with. AI capability has surpassed just answering questions or reviewing documents. The leading solutions can now run workflows end-to-end, where the work actually gets done (e.g., carrier portals, AMSs, employee navigator, etc.).
What is the baseline for how long a workflow takes? Measuring current turn-around time is not just useful for measuring ROI, but it should also help you determine whether or not something should even be automated in v1. It's very easy to get excited about cool flashy "nice-to-haves" than use cases that'll actually move the needle.
What inputs does the system need to be reliable? AI models have become a commodity. What differentiates a great AI system vs. just using a generic LLM is a combination of receiving rich data inputs + designing a memory layer + giving the agent access to this unlimited memory. This is what brings the output accuracy to 98%.
What about the 2%? The best-designed AI systems asks users questions if unsure, so hallucinations are effectively reduced to 0.
What would success let the team do? The answer should be more groups served, improved customer service, increased quality of work, re-allocating producer time to sourcing more clients. “Hours saved” "x% less time per task" is pointless if teams are not intentionally re-allocating the new capacity towards driving more value.

Closing remarks

The window is still open. 99% brokerages still have not rebuilt their operating model around AI. But the firms that are moving first are already setting a higher bar for what a service team can carry and what a client should expect.

For smaller brokerages, the scale is obviously different. The answer isn't necessarily hiring McKinsey or subscribe to Claude.

If you haven't already, subscribe to this newsletter for more factual looks at how technology is changing the brokerage business.

Sources

← All editions