Extract the Declination Reason From the Email. Never Ask an Underwriter the Same Question Twice.

Sivan Iram spent close to a decade as a software engineer before deliberately shifting toward the business side of technology, coming to the US roughly 11 years ago for an MBA and working across digital transformation projects in automotive, aerospace, supply chain, and manufacturing. When he decided to start his own company, he wasn’t attached to a particular industry — he was looking for a large, growing market with a genuine, technology-solvable pain point.

Insurance found him through his brother, who’d built high growth at Bay (a prior insurtech venture) — conversations that led Sivan into dozens of broker and underwriter meetings, seeing the industry through their eyes before committing to build anything. What he settled on: middle-market commercial insurance, deliberately distinct from small commercial, where humans remain structurally necessary in the transaction — agents and brokers representing the insured, underwriters doing genuine actuarial judgment work — rather than a segment ripe for full automation and disintermediation.

In Episode 105 of InsurTechTalk, Sivan and I covered what Capitola actually is (not an MGA, not an agency management system replacement), how the company became a digital wholesaler almost as a natural extension of its core product, and a genuinely sharp example of GPT extracting underwriting intelligence from ordinary email correspondence.

About Sivan Iram

Sivan Iram is the CEO and co-founder of Capitola, a software platform serving middle-market commercial insurance brokers and underwriters, founded in 2021 alongside co-founders Amit Bintan (from Google) and Noam Rosenberg (leading R&D from Israel, previously at Optimus). Capitola raised its seed round from Lightspeed Venture Partners and closed a $16 million Series A led by Munich Re Ventures roughly six months before this recording, with Lightspeed participating again. The company positions itself as a smart digital marketplace connecting middle-market commercial insurance buyers (via agents and brokers) with sellers (underwriters at carriers and MGAs), and operates as a licensed digital wholesaler in over 85% of the country.

Not an MGA, Not an AMS Replacement — A Placement Management System

Sivan was precise about what Capitola is and isn’t, given how much confusion exists around adjacent insurtech categories:

• Capitola is a software company, not an MGA — it doesn’t take on underwriting risk itself
• It doesn’t replace an agency’s existing agency management system (Applied Epic, Vertafore AMS360/Sagitta, and Salesforce-based next-generation platforms) — it integrates with them, pulling in book-of-business and carrier data via hooks into the AMS
• Functionally, it’s best understood as a placement management system: brokers and account managers log in to see their entire book ready for placement, gather application documents and loss runs, get data-driven market recommendations, manage back-and-forth negotiation with carriers in a structured workflow rather than through Outlook, and generate client-facing proposals with a single click — with information feeding back into the AMS for billing

The stated design philosophy: automate everything that doesn’t require actual insurance expertise (manual data entry, repetitive administrative tasks), and free brokers and underwriters to focus on the parts of the job technology genuinely can’t replace — relationship-building, negotiation, and risk-specific judgment.

Becoming a Digital Wholesaler, Almost Accidentally

This was the most structurally interesting part of the conversation, and it illustrates how a software company can end up performing a wholesale distribution function without ever intending to compete with traditional wholesalers.

Capitola’s core recommendation engine points brokers toward the best-fit carrier for a given risk. Most of the time, that recommendation lands on a carrier the broker’s agency already has a direct appointment with — and Capitola stays out of that relationship entirely. But when the best-fit capacity provider for a specific risk is a carrier the agency doesn’t have a direct appointment with, Capitola steps in as the wholesaler itself — filing taxes and fees, and placing the business through its own wholesale operation, licensed across the majority of US states with a goal of full 51-state coverage.

Sivan’s framing of the underlying philosophy: Capitola’s mission is to put the best available capacity at every broker’s fingertips, which sometimes structurally requires becoming the wholesale intermediary rather than staying purely a software layer sitting outside the transaction — but only when there’s genuine value to add, never inserting itself between a broker and a carrier relationship that already exists directly.

GPT for Extracting Underwriting Intelligence From Email

Sivan’s framing of AI in Capitola’s product is worth quoting almost directly: “AI is not what you do, it’s how you do it” — the company isn’t “an AI company” in the marketing sense; AI is a building methodology serving a specific mission of connecting the industry more efficiently.

His concrete case for why insurance specifically benefits from large language models, as distinct from a lot of AI hype elsewhere: insurance runs almost entirely on textual information — applications, quotes, binders, policies, and email correspondence. A model built for imaging tasks (say, identifying cancerous cells in an MRI) would be useless here; a model that thrives on ingesting large volumes of unstructured text is exactly matched to how this industry actually generates and stores its working knowledge.

The specific example he gave is genuinely sharp: when an underwriter declines a risk over email — something that happens constantly, informally, and is almost never captured anywhere structured — Capitola extracts the declination reason directly from that email correspondence and feeds it into its recommendation engine. His concrete illustration: an underwriter declining a property risk specifically because “we don’t underwrite anything built prior to 1946.” The next time a similar risk comes through the platform, Capitola can proactively flag that history to the broker — not blocking them from still trying that market, but surfacing institutional knowledge that would otherwise live only in one underwriter’s inbox and one broker’s memory, if it was remembered at all.

That’s the pattern worth generalizing: the AI isn’t making underwriting decisions — it’s compiling and surfacing dispersed, previously unstructured institutional knowledge so a human can make a better-informed decision faster, which is precisely the kind of “decision-making at scale” application that tends to hold up better than more speculative AI use cases.

Munich Re Ventures as a Strategic Partner, Not a Competitor

Sivan was direct about how the relationship with Munich Re Ventures functions differently from a typical strategic corporate investment. Munich Re Ventures brings genuine value by helping Capitola expand its carrier network for the wholesale offering — a digital wholesaler is only as strong as its carrier relationships — but critically, Munich Re doesn’t operate as an active player within Capitola’s own marketplace. Sivan noted this distinction explicitly: some companies with strategic reinsurance investors face concern that the investor is also a competing actor on their own platform; Capitola hasn’t experienced that friction, since Munich Re Ventures functions purely as a capital and network partner rather than a marketplace participant.

Advice: Surround Yourself With the Right People, and Learn to Ask

Asked for closing advice, Sivan credited a college professor’s line that’s guided his approach across every industry transition in his career: you don’t need to know the answer, you need to know who to ask. His practical application, repeated across moves from software engineering into digital transformation consulting and then into commercial insurance specifically: surrounding himself with genuinely knowledgeable people who buy into the mission, at every step, has been the consistent differentiator — more so than any individual technical or domain expertise he brought in himself.

Key Takeaways

• Middle-market commercial insurance deliberately retains human intermediaries (agents, brokers, underwriters) by structural necessity — Capitola’s thesis is built on augmenting that workflow, not automating humans out of it
• A placement management system integrating with, rather than replacing, existing agency management systems avoids forcing agencies into a disruptive platform migration to get value
• A software company’s recommendation engine can organically evolve into a wholesale distribution function whenever the best-fit carrier for a specific risk falls outside a broker’s existing direct appointments
• Insurance is a genuinely strong fit for large language models specifically because the industry’s working knowledge is almost entirely textual — applications, correspondence, and policy documents — unlike domains requiring computer vision or other non-text modalities
• Extracting informal, previously unstructured knowledge (like a declination reason buried in an email) and feeding it back into a recommendation engine is a concrete, high-value AI application that augments human judgment rather than replacing it
• A strategic reinsurance investor can add genuine value (carrier network access) without creating competitive conflict, provided the investor doesn’t also operate as a participant within the startup’s own marketplace
• Career and industry transitions are best navigated by surrounding yourself with people who already have the expertise you lack, rather than trying to build that expertise entirely independently before acting