The opinions expressed by Digital Journal contributors are their own.
Underflow’s CEO and former investment banker says the industry has spent two decades automating the wrong thing. Its AI-powered alternative is turning heads.
The commercial insurance industry has no shortage of technology vendors promising to modernize underwriting. Over the past decade, insurtech startups have raised billions of dollars on the premise that software can make the process faster, cheaper and more accurate. And yet, by most measures, the basic rendering workflow has barely changed.
Ola Kolade has a theory as to why:
“The industry kept trying to automate the container instead of automating the job,” Kolade said. “You can digitize a filing cabinet. You can build a prettier inbox. But if the underwriter has to open every attachment, read every page, write every data point into a system, and draft every follow-up email by hand, you haven’t actually automated anything. You’ve just given them a prettier desk to do the same manual work.”
It’s a sharp critique, and one that’s gaining traction. Kolade, CEO and co-founder of Underflow Inc., a San Francisco-based startup that has raised $3 million in seed funding, is building what he calls an autonomous signature assistant. The product, called Aurora, doesn’t try to improve existing workflows. He tries to completely eliminate the manual parts of it.
The numbers behind the problem
The degree of inefficiency that Kolade describes is well documented. Capgemini Global Property and Casualty Insurance Report 2024 found that 41 to 43 percent of commercial underwriters’ time is spent on administrative activities, including data entry and record keeping. Only about a third of their time goes to core underwriting functions like risk assessment and premium calculation.
Accenture Research paints an even starker picture, finding that the average commercial entrepreneur spends roughly 70 percent of his time on non-signature activities. signature estimated the cumulative cost of this inefficiency to $85 billion to $160 billion over five years across the industry.
These are not obscure statistics in the world of insurance. Freight forwarders, general managing agents and wholesalers quote them regularly. The question that has frustrated the industry for years is what to do about it.
A stranger’s diagnosis
Kolade did not appear in the security ranks. A Canadian citizen who studied economics at McGill University, he began his career in investment banking at National Bank Financial in Toronto, where he worked on more than ten transactions across the technology sector. He later moved to San Francisco to lead strategic finance at Faire, one of North America’s largest wholesale marketswhere he led the top-level business unit responsible for the platform’s largest retail clients.
Prior to that, at Power, Kolade negotiated $10 million in debt financing, built integrated three-statement financial models and KPI dashboards supporting major capital raises, and led a strategic advisory board that coordinated quarterly meetings with executives across the organization.
His introduction to commercial insurance came through Faire, where he observed the risk management processes of thousands of independent businesses on the platform. He began studying the commercial insurance distribution chain, and what he found struck him as a fundamental discrepancy between the sophistication of the people in the industry and the primitiveness of their tools.
“I came from an environment where we used machine learning to match retailers with real-time inventory,” Kolade said. “And then I looked at how an insurance submission goes from a broker to a carrier, and it was email threads, PDF attachments, and manual rekeying in legacy systems. These are great professionals working with tools from 2002.”
The understanding gap
Kolade published his analysis of the problem in December 2025 at Rough Notes, one of the oldest insurance trade publications in the United States, founded in 1878. The article, which examined why traditional software has failed to transform underwriting workflows, argued that previous solutions addressed the wrong layer of the problem.
The bottom line, Kolade wrote, is not that documents are difficult to store or organize. It is that they are difficult to understand. A typical commercial insurance submission includes ACORD applications, losses from previous carriers, schedules of values and supplemental forms. Each document contains data that must be interpreted in the context of a specific line of business, correlated with operator requirements, and evaluated for completeness and consistency.
“Optical character recognition can read the text on a page,” Kolade said. “It can’t tell you that the loss period is two years short of what the carrier requires, or that the type of construction listed on one form contradicts the information on another. That’s the gap. Not the reading. The understanding.”
How Aurora approaches the problem
Aurora, the product Kolade built through Underflow, connects directly to a firm’s existing email environment. When a submission arrives, Aurora reads the email and any attachments, extracts all relevant data, and structures it into a unified record. It then performs a detailed gap analysis, identifying not only that a submission is incomplete, but specifying exactly what is missing and why.
The system then drafts and sends follow-up communications to the broker, either under its own identity or through the firm’s email accounts. If a response does not arrive within a certain period, it continues again. When the file is complete, Aurora packages it and directs it to the designated provider.
Critically, Aurora also performs functions that go beyond ingest processing. He writes preliminary risk assessments, points out inconsistencies across documents, identifies potential coverage opportunities, and enriches submissions with research from public data sources, including property records, business records, and corporate profiles.
“We’re not building a faster version of the existing process,” Kolade said. “We’re building a system that delivers a better result. The insurer should get a file that’s not just complete, but contextualized, with analytics already done and external data already integrated. That’s the standard Aurora is built to meet.”
Retirement Rock
Kolade’s thesis is sharpened by a demographic reality the industry has been chasing for years. Roughly a quarter of the commercial insurance workforce is 55 or older, and the institutional knowledge they carry, including knowing carrier appetite, risk model recognition and relationship context built over decades, is not something that can be transferred through a training manual.
“The industry is losing 20 and 30 years of experience every time someone retires,” Kolade said. “You can’t buy your way out of that. The math doesn’t work. What you can do is capture the repeatable parts of that expertise in a system that scales, so your remaining professionals can focus on the judgment calls that actually require human experience.”
Skepticism and the way forward
The insurance industry’s relationship with technology has historically been cautious. Adoption cycles are long, switching costs are high, and trust in new systems is hard-won. Kolade says he anticipated this.
“We built Aurora to work inside email because that’s where insurance happens,” he said. “We’re not asking anyone to change the way they send submissions. We’re not asking for a new portal or a new login. The broker sends the email the way they’ve always done it. The difference is what happens when it arrives.”
Whether Underflow can overcome the industry’s institutional inertia is the key question facing the company. But Kolade’s framing of the problem, distinguishing between document management and document understanding, has resonated with practitioners who have seen previous waves of technology fall short of their promises.
As Kolade wrote in Rough Notes: “Speed has always determined who gets business. Now it will determine who stays in business.” The industry, it seems, is listening.
