HUB International just rolled out Claude across 20,000 employees and reported 85% productivity gains. Two and a half hours saved per employee per week. Ninety percent user satisfaction.
The agencies that read that and think "that's a big company story" are the ones that will be rebuilding in three years.
The number that matters isn't the headcount. It's the measurement system behind it.
The Main Insight
Why Deploying AI and Measuring AI Are Two Completely Different Things
There is a version of this that most agency owners are living right now and don't realize is a problem.
They have AI. It's doing things. They can feel it helping. But if you asked them what their AI produced last week, how many leads it qualified, how many renewals it touched, how many hours it actually saved, and whether the quality of its outputs improved or degraded over the last thirty days, most of them couldn't tell you.
That's not an AI system. That's an AI experiment that nobody is running.
Here's what that costs.
AI that isn't measured doesn't improve. It drifts. The prompt that worked in January gets stale by March. The workflow that was producing clean outputs starts producing mediocre ones. Nobody catches it because nobody is looking. And six months later the agency owner is telling someone the AI "kind of works" without being able to explain why it used to feel better.
The agencies that are compounding on AI right now have one thing in common. They treat their AI outputs the same way they treat their producers. They grade the work. They track performance over time. They know when something is slipping before it becomes a problem.
HUB International didn't report 85% productivity gains because they deployed Claude. They reported 85% productivity gains because they measured it. They tracked adoption. They monitored outputs. They held the system accountable to benchmarks. The measurement is what turned a deployment into a result.
That same logic applies to a two-person agency in Florida as much as it applies to a 20,000-person brokerage.
The agencies that figure this out now are the ones building compounding operations. The agencies that don't will spend 2027 trying to understand why their AI investment never paid off.
Stories Worth Your Attention
HUB International rolls out Claude to 20,000 employees and reports 85% productivity gains.
One of the largest insurance brokerages in the country just completed a full enterprise rollout of Anthropic's Claude platform. Early results: 85% productivity lift, 2.5 hours saved per employee per week, and user satisfaction above 90%. HUB didn't get those numbers from the tool. They got them from the accountability layer they built around the tool. Every independent agency owner reading this can apply the same principle at any scale. You don't need 20,000 employees to measure whether your AI is working. You need a system that tells you what it produced and whether that output was good.
Allianz and Anthropic just announced a global partnership to build agentic AI for end-to-end insurance operations.
Allianz is deploying custom Claude agents to orchestrate multi-step workflows across claims processing, underwriting, and intake documentation. The stated goal: fewer manual steps, faster first payments, better client experience at the moments that matter most. The carriers are building this now. The large brokerages are building this now. The gap between agencies that have agentic operations and agencies that don't is going to be visible to carriers, visible to clients, and visible in your retention numbers within two years. The agencies that start building the foundation today are the ones that will be ready when that gap becomes the difference between preferred appointments and losing them.
Consumer support for AI in insurance nearly doubled in a single year.
Insurity's 2026 AI in Insurance Report found that 39% of consumers now support their insurer using AI to improve services. In 2025 that number was 20%. Consumer resistance is softening fast. But the most important number in the report is where the comfort line sits. 46% of consumers are comfortable with AI generating a quote. Only 16% are comfortable with AI canceling or renewing a policy without human involvement. The window between AI-assisted and AI-replaced is exactly where independent agencies win. Consumers want the speed and accuracy AI delivers on the back end. They still want a human on the relationship. The agencies that deploy AI underneath the client experience while keeping the human on top are the ones consumers will choose in 2026 and beyond. That is a structural advantage national carriers cannot easily replicate.
Early Mover Opportunity
The generative AI market in insurance hit $1.11 billion in 2025. By 2035 that number is projected to reach $14.35 billion. A 29% compound annual growth rate for a decade straight.
That is not a prediction. That is capital that has already been committed based on certainty that this is where the industry is going.
By late 2026 more than 35% of insurers will deploy AI agents across at least three core functions. Claims processing 75% faster. Operating costs down 30 to 40 percent.
The agencies building now are not chasing a trend. They are getting in front of a market that already decided.
A 29% CAGR compounding for ten years does not produce a slightly better agency. It produces a fundamentally different kind of agency. The agencies inside that compounding curve and the agencies outside it will not be competing for the same clients by 2030. They will be operating in different markets entirely.
The question isn't whether to build. The question is whether you start measuring now or start rebuilding later.
Tool Spotlight
How to Know If Your AI Is Actually Working
Skip the feature comparisons this week. Here is the only framework that matters right now for agency owners who already have AI deployed.
Can you answer these five questions about your AI stack?
What did your AI produce last week, specifically? Not "it helped with emails." How many outputs. What type. What volume.
Did output quality improve or decline over the last thirty days? If you don't have a way to answer this, you don't have a measurement system.
What is your AI costing you per output? Token cost per lead response, per renewal reminder, per triage decision. If you don't know this number you are flying blind on ROI.
When your AI produces a bad output, how do you find out? If the answer is "a client complains" you have a quality control problem.
Which of your AI workflows is performing best and which is underperforming? If you can't rank them you can't improve them.
The agencies that can answer all five are compounding. The agencies that can't answer any of them are running an experiment they aren't monitoring.
This is exactly the problem Signal Social was built to solve. Every AI output scored. Every agent ranked. Every workflow measured against benchmarks. The system tells you what's working before you have to guess.
The agencies that win this decade will be the ones that measure early and compound on what works.
If you want to find out where your agency's AI stack actually stands, the free Agency AI Audit is at closemodeai.com. Eight minutes. Forty questions. No sales call.
All Sauce. No Filler. Just Signal.
— Rhett
CloseMode AI | closemodeai.com