An uncomfortable truth about distribution is emerging: as carriers and brokers automate the wrong things, they ignore what clients actually value.

The insurance industry has spent years obsessing over submission speed. Digitize the intake. Extract the data. Route it faster. Get the quote out quicker.

But at InsurTech NY’s two-day Spring Conference in New York, a panel on “Distribution Beyond the Submission” challenged that premise.

“Do clients care about faster submissions? Probably not,” said Raj Kalahasthi, founder and CEO of Catalyx Advisors. “If you ask a CFO or a risk manager or a business owner, the most value they get from a broker is when they’re present at the time of need. Did they fight for my claim? Did they help me find a risk I didn’t find otherwise?”

The panel — moderated by Heather Turner, program manager at ReSource Pro and contributing editor at The Insurance Lead — featured perspectives from across the distribution ecosystem: Pilar Lorenzo, intermediaries business unit head at Chubb; Sandeep Haridas, EVP and insurance business head at IntellectAI; and Garrett Koehn of IA Seed Ventures.

The main takeaway: the industry is automating backend reconciliation and margin efficiency while largely downplaying the upfront advisory work that actually creates client value.

Where Automation Actually Matters

Haridas pushed back on the notion that speed doesn’t matter, at least for certain segments that depend on efficiency for fast, reliable data extraction.

“A faster submission is required in the micro-commercial space, in small commercial, even in large commercial,” he argued.

Ten years ago, as an underwriter, Haridas would send submissions to rating specialists who wouldn’t look at them for at least a week. “During these five days, if the agent or broker called me asking about the status, I would not answer the call because I had no idea what the submission was.”

Today, technology can extract and surface submission data in 15-to-20 minutes, Haridas noted. “That means as an underwriter, if I had five days to put a quote together before, now I have 10. The quality of my decision-making, the quality of my service back to my producer is going to be a lot better.”

That additional time enables different conversations — not just about large losses, but about frequency issues, near-bankruptcy situations that didn’t show up in records, and secured debt problems. Better data extraction means better decisions and better service.

Chubb’s Lorenzo warned against a romance with automation — particularly in claims. She cited a large carrier that fully automated its claims process and watched its Net Promoter Score, an online tool that measures customer satisfaction, plummet within 10 days.

“Clients and brokers were complaining about the lack of human touch, the fact that the process was being dehumanized,” Lorenzo said. “When someone has a claim — a death claim, loss of a home, a vehicle accident — that human interaction is very important.”

Automation should be a copilot, she argued, not a replacement for judgment and empathy at critical moments.

The Data Problem Nobody Wants to Discuss

Koehn, formerly president of brokerage at CRC Group, described the data challenge bluntly: agency management systems contain almost nothing useful for underwriting.

“You have an address, you’ve got a premium. You don’t know if it’s work comp, how many employees there are, what class codes they’re in,” he said. “There is effectively no useful data that you can use with clients and anybody else.”

At CRC, they had to aggregate useful data manually for years. Now, AI companies are getting better at scraping submissions and extracting that data automatically. But even with good data, using it remains cumbersome.

“What I used to do was call up Scott, send him an email: ‘Scott, can you run this report on lawyers?’ We’d go back and forth; it’d be wrong; we’d change it. Two weeks later, three weeks later, maybe we might finally have the data I needed.”

Although language models sitting on top of data systems can now return that information instantly, Kalahasthi pointed to a deeper data gap: decision context.

“You got the data, fine. But how do you capture the decision context? Why did you go with this carrier? Why did this underwriter approve? What really happened in that relationship?” That context lives in someone’s head, Kalahasthi said, citing the advisor, the producer, and the service person as examples. Capturing the context of those relationships typically goes unsolved.

The Disintermediation Question

Turner posed an essential question: “With carriers gaining direct customer access through technology and AI automating distributor work, what’s the future role of intermediaries?”

Koehn wasn’t concerned about wholesalers. “There’s a big moat on wholesale. If you’re starting up a new facility and you need to get to 60,000 agents immediately, all you have to do is give an appointment to RT, and they can do that for you.”

But he acknowledged compression at other layers. Policy checking used to cost $100 per unit, then dropped to $10 when outsourced to the Philippines and China. “Now AI is doing it for a dollar a unit.”

Business process outsourcing represents the first wave of displacement. The next segment: mid-level people doing back-and-forth work who don’t control client relationships and aren’t producers. “AI is going to be able to do a better job than a lot of those folks over time.”

Haridas was more direct: “The jobs that are going to be lost are going to be lost. People whose jobs involve taking data from one place and putting it into another system — those roles will not exist and should not exist in the future.”

However, anyone who owns relationships — retail agents and brokers, wholesale brokers connected to retail agencies, underwriters connected to producers — will have roles in the new economy. For those brokers, the threat isn’t in automation, but in the nuanced use of it. “You’re not going to be threatened by AI,” Haridas said, channeling a familiar observation. “You’re going to be threatened by somebody who’s going to use AI.”

Lorenzo mentioned that governance and compliance will play essential roles in preserving human involvement where it matters. “There needs to be a balance as long as there are clear guardrails around AI initiatives — compliance management to put limits around digital efficiencies and what needs to happen around transactional things to allow humans to really focus on the complexity of human relations and products that need human mind involvement.”

Build vs. Buy — and Who Should Own What

Koehn offered pointed advice on technology strategy: stop trying to build everything internally.

“There are a lot of IT departments with really big AI beer muscles right now,” he said. Whenever there’s a large scalable space like insurance, startups will enter with better people and more capital than internal IT departments can match. “You’re eventually going to lose if you don’t lose right away because you’re not going to be able to monetize the products you’re building beyond your own company.”

Let IT departments stick to keeping the business running. Partner with startups doing quality AI work in specific domains, Koehn advised.

Kalahasthi agreed but added a critical qualifier. “As a technology leader, you need to own the orchestration layer. You need to own and be able to plug-and-play. Your vendor’s roadmap cannot be your roadmap. You own the orchestration layer and use different technologies or vendors to plug-and-play to achieve your client journey.”

He outlined three pillars for survival:

• Get clarity on the future optimal operating model
• Demonstrate the courage to rewire that operating model rather than just slapping AI onto existing processes
• Claim ownership of the orchestration layer, while letting vendors handle components

There is yet some complexity to that simple advice. The operating model question varies by business line. At Baldwin Group, where Kalahasthi previously served as CTO, the strategy was human-led and AI-assisted on the commercial side, AI-led with humans in the loop on the personal lines side.

TIL Takeaway: The insurance distribution model is shifting from transaction speed to relationship depth. While AI automation eliminates low-value work like data entry and policy checking (dropping costs from $100 to $1 per unit), client value concentrates in advisory capabilities: claims advocacy, risk discovery, and decision context that agency management systems can’t capture. The disintermediation threat isn’t AI itself—it’s competitors who use AI to free capacity for advisory work while maintaining the orchestration layer. Successful brokers and carriers will operate human-led with AI assistance in complex commercial lines, and AI-led with human oversight in transactional personal lines. The jobs disappearing are BPO roles and CSR functions that move data between systems; the jobs surviving are relationship owners who deliver expertise at moments of need. Key execution principle: own your orchestration layer, let vendors handle components, and rewire operating models rather than automating broken processes.