Flow’s Broker-in-the-Loop Model: Why AI Without Human Supervision Misses the Point

Flow’s Broker-in-the-Loop Model: Why AI Without Human Supervision Misses the Point

In a recent episode of Category Visionaries, Sivan Iram, CEO and Founder of Flow, explained why his AI-powered insurance wholesaler refuses to go fully automated.

The technology exists to let AI handle submissions end-to-end. Structure the data, match it to carrier appetites, place the business, gather quotes, compare coverage, send proposals. Flow’s AI agents can do all of it. And in an industry where speed equals competitive advantage, cutting out the human step would make them even faster.

But they don’t. Every single submission gets human supervision. Every quote, every placement decision, every client interaction has a broker in the loop.

This isn’t a temporary compromise while the AI gets better. It’s the permanent architecture. And it’s why Flow is tripling revenue every month while competitors debate whether to automate away their workforce or stick with pure labor.

The False Choice

Most companies implementing AI face what seems like a binary choice: automate everything and become radically efficient, or keep humans in the process and accept higher costs.

Flow rejected both options. Instead, they asked a better question: which parts of our process require human judgment, and which parts are just expensive busy work?

“We take in submissions via email with attachments. And we employ technology to be really efficient with the way that we place. We structure that data with AI agents,” Sivan explains. “The core technology for Flow is a system of AI agents that help our brokers every step of the way. From structuring the submission, recognizing what carriers have that in their appetite. So to do risk appetite matching place business in the right way.”

The AI handles the mechanical work—extracting data from PDFs, categorizing risks, identifying which carriers to approach, even drafting initial submissions. This is work that doesn’t require expertise. It just requires accuracy, consistency, and speed.

But here’s the crucial part: “We do all that with AI agents, but we always have a broker in the loop, we always have human supervision.”

What Humans Are Actually Good At

The reason brokers hire wholesalers isn’t to get their submissions structured or their data entered. They can do that themselves. They hire wholesalers for expertise, market knowledge, and relationships.

When Flow first launched, they learned this lesson the hard way. They built a self-service platform where brokers could log in and place business directly with carriers. No human intermediary needed. Just pure efficiency.

It failed. “The bigger issue was that they were looking to us as their wholesaler, not only to get connected with our markets, which they could on the platform, but they needed our expertise, they needed our guidance, they needed our advice and insights,” Sivan recalls.

Brokers don’t want to be more efficient at doing the wholesaler’s job. They want the wholesaler to do the job better. The value lives in judgment, not in process execution.

So Flow rebuilt around this insight. AI executes the process. Humans provide the judgment. The combination delivers both speed and expertise—something neither pure automation nor traditional labor can match.

The Hybrid Advantage

Flow’s broker-in-the-loop model creates advantages pure automation can’t replicate.

First, it maintains expertise. When submissions have unusual characteristics or fall outside standard parameters, the human broker catches it. When carrier appetite shifts, the broker knows it. When clients need guidance, the broker provides it.

Second, it preserves carrier relationships. Insurance underwriters want to work with people they trust, not algorithms. The AI drafts submissions, but broker relationships get them approved on favorable terms.

Third, it enables flexibility. “Sometimes it’s via APIs if the carriers have it. Sometimes it’s going to be with all traditional email. If the coverage is great or the quote can be really competitive, the market could be really competitive on pricing, but they don’t have an API. We definitely do not want to exclude them from the placement,” Sivan explains.

Strategic decisions about which carriers to approach require human judgment.

The Economics That Make It Work

The broker-in-the-loop model only works if economics justify keeping humans involved. Flow’s AI changes those economics dramatically.

Traditional wholesalers face brutal trade-offs on small accounts. “If you make 20%, you make 10% commission on those deals, you know, and it’s a $10,000 deal. You can’t really spend a lot of time on it if you’re a traditional player,” Sivan notes.

But with AI handling mechanical work, human broker time investment drops dramatically. They’re not spending hours on data entry or submission formatting. They’re spending minutes on oversight and judgment calls.

This makes small accounts profitable while maintaining quality. “For us it’s very different. We’re able to place business with our carriers very quickly and allows us to really invest a lot in every size account.”

The hybrid model turns what seemed like a cost into an advantage: providing expertise at scale.

The Pattern for Other Industries

Flow’s approach reveals something important: the goal isn’t to eliminate humans or keep them doing everything. It’s to redesign the division of labor.

Most AI implementations fail in two ways. They try to fully automate processes requiring human judgment, leading to poor outcomes. Or they slightly augment existing processes without rethinking workflows, leading to marginal improvements that don’t justify investment.

Flow completely redesigned their workflow around one principle: AI does everything that doesn’t require expertise, humans provide all the judgment and relationship management customers value.

This applies to any professional services business where part of the process is mechanical, part requires expertise, clients pay for outcomes not process execution, and speed matters but can’t sacrifice quality.

The key: be honest about what clients hire you for. If they hire you for expertise and outcomes, automate everything else. In most high-value B2B services, clients pay for judgment—they just don’t want to wait weeks for it.

Why This Matters

Flow’s growth validates the model. “In the past nine months, we’ve been tripling our top line every single month. We’ve been tripling our submission volume.”

This isn’t growth from discounting or niche targeting. It’s from being demonstrably better: faster than traditional wholesalers, more expert than pure automation, more scalable than labor-intensive services.

The broker-in-the-loop model makes that combination work. Remove the human, lose the expertise. Remove the AI, lose the speed and economics. Together, they create something neither could achieve alone.

For founders implementing AI, Flow’s lesson is clear: don’t automate for automation’s sake. Automate work that doesn’t require human judgment, then deploy that judgment where it creates value. The goal isn’t efficiency or expertise—it’s both.