The real reinvention of this industry over the coming decade will come from learning from the best underwriters, combining and scaling that expertise through technology, and applying it to real-time portfolio management.
“I compare [Fortune] to one of those raging rivers, which in flood overflows the plains, sweeping away trees and buildings, bearing away the soil from place to place; everything flies before it, all yield to its violence, without being able in any way to withstand it; and yet, though its nature be such, it does not follow therefore that men, when the weather becomes fair, shall not make provision, both with defenses and barriers, in such a manner that, rising again, the waters may pass by canal, and their force be neither so unrestrained nor so dangerous. So it happens with Fortune, who shows her power where valour has not prepared to resist her, and thither she turns her forces where she knows that barriers and defences have not been raised to constrain her.” - NICCOLÒ MACHIAVELLI, THE PRINCE
When people ask what differentiates our approach at Vivere, I often start with Machiavelli. This strikes most of them as strange. What does a 500-year-old text, best known for lines like “it is better to be feared than loved,” have to do with AI-native insurance? (As an interesting aside, scholars still debate whether The Prince is sincere or a veiled satire of the oppression of Machiavelli’s era.) To answer that, we first need to understand what makes a good underwriter, and from there how we see the technology opportunity in insurance.
Broadly, an expert underwriter operates at three levels. You can think of seniority as the highest level an underwriter is able to reach.
━━
Level 1: Operational
Receive the submission by email, go through the attachments and the body, assess appetite and pricing against a clear set of guidelines, fill in the rater from the information provided, and apply subjectivities or endorsements according to a deterministic set of rules.
This is the level of the underwriting assistant or junior underwriter, and it is already automatable with foundation models. It is where most of the interest in “AI transformation” has concentrated, and where the insurtech SaaS offerings compete on a fairly commoditized set of functionality. That’s understandable; it’s the lowest-hanging fruit. Most carriers and MGAs are held back from even this layer of automation by legacy systems, archaic processes, and poorly integrated workflows. It’s common to hear of a carrier that buys a modern AI product for submission intake, discovers it can’t be connected to the rating tool or policy admin system, and ends up with underwriters keeping the shiny new toy open on one monitor while doing, on the other, the same data entry it was meant to eliminate.
At Vivere, we’re building the entire technology and operational platform from scratch, designed from day one around automation and what’s possible in the age of AI. That unified approach lets us find bottlenecks and restructure the end-to-end process to remove them. It has also let us identify the core abstractions underlying every line of insurance and build a set of AI-native building blocks for launching new lines with unusual speed. Standing up a new line traditionally takes nine to twelve months of technology and operations work. By configuring our platform’s capabilities — AI document extraction, agentic research, rating, document templating, and evals — we’ve lately been able to stand up a new, highly efficient program in about two weeks. That velocity lets us launch and support many programs in parallel with a lean engineering team.
Our lack of operational and technical debt is a real advantage, and it lets us move and scale quickly. But this technology is too commoditized to be a long-term moat.
━━
Level 2: Risk-driven
This is where a strong senior underwriter operates: drawing on judgment and instinct, honed over years of seeing similar risks and their outcomes, to decide how a risk should be priced (not so low that we lose money, not so high that the insured scoffs) and how coverage should be tailored (broad enough to serve the insured, narrow enough to protect the carrier from unnecessary exposure). It’s a delicate balance, and in many ways more art than science. What makes it harder is that underwriters work from a very limited set of information; only a small fraction of what you’d want to know about a risk ever makes it into the application. We augment this with third-party data and whatever can be scraped from the internet, but gut feeling remains a large part of what makes a good underwriter good. The loss runs are clean, but the broker’s cover note is a little too breezy about a change in operations two years ago. A management liability applicant has a founder with a history of being sued, and that fact appears nowhere in the application.
A seasoned underwriter gets a hunch and pulls on the thread — through their own research or by going back to the broker — and comes back with a decision: whether to write the account at all, at what price, and with which exclusions, sublimits, or warranties bolted on to make the risk acceptable. None of this is in the guidelines, because it can’t be. The world is too complex. You can’t prompt-engineer your way to intuition.
What you can do is collect data on how underwriters approach a risk: what they spend time looking at, where they dig in or pull up a spreadsheet, what they ask the broker, and what they finally quote.
That is exactly the kind of specialized, high-skill training data the frontier labs are paying enormous sums for across every knowledge-work vertical. No one is collecting it for specialty underwriting. The work happens inside carriers and MGAs that have no interest in instrumenting it, and no ability to if they wanted to, on systems built to record nothing beyond the final answer.
Once we collect that data and train our own models on it, it becomes our moat. Open-weight models have grown powerful enough over the past year that raw intelligence is no longer the constraint. The bottleneck to expert-level performance is data: enough of it to replicate the reps an underwriter accumulates over a career of taking shots and seeing what binds and what loses money.
━━
Level 3: Portfolio-driven
This is the level at which a chief underwriting officer must operate. A CUO isn’t underwriting individual accounts; they’re managing a book. How concentrated is the exposure? How correlated? How is the mix shifting as the market softens, and how do evolving trends and volatility change the risk landscape?
Today this is done with poor, stale information. The CUO reviews a monthly or quarterly report assembled by hand. Drift in the book is discovered a quarter after it happens, by which point the accounts are bound for a year.
Visibility into aggregate risk is worse. The industry manages correlation through a handful of explicit dimensions that are easy to track: zip code, industry, state. It’s the drunk searching for his keys under the streetlight, because that’s where the light is. Correlation in a book can be far more insidious.
Cyber is the obvious example. Log4j, MOVEit, and the CrowdStrike outage each hit thousands of insureds across every industry code; the only thing they had in common was a piece of software in their stack. In D&O, companies that went public via SPAC were sued at multiples of the rate of traditional IPOs regardless of sector, and more broadly, IPO cohorts from a frothy year carry correlated exposure because they’re priced against the same market and tend to rise or fall together. In EPL, Illinois’s biometric privacy statute produced massive, correlated claims against employers using fingerprint timeclocks — an exposure defined by a piece of HR hardware. There are too many dimensions along which to slice a portfolio for any human to assess the probability and impact of every scenario, let alone in real time.
This is where we think technology can give underwriting leaders a degree of insight into their books that no person could achieve alone.
Our underwriting platform, Fawkes, holds a far richer description of each risk than any bordereau: everything extracted from the submission, everything the underwriter went and looked up, and everything we could enrich it with from third-party sources. That makes correlation a question you can ask of the book at any time rather than a set of dimensions fixed in advance. When a new liability theory emerges, or a fire burns through an area built with a flawed construction method, we can infer the link from our claims data and readjust our risk assessment proactively, instead of waiting for folk knowledge to spread by word of mouth.
Second, we — a human, or eventually an army of agents — can simulate scenarios and stress-test the book against every plausible external event, to understand not just the marginal risk distributions but the joint ones. And we can run that evaluation at the moment of each quote. If the book already carries heavy exposure to some latent feature, the next account sharing it is worth less to the portfolio than the last one was, and the price and terms should say so. Today that adjustment is crude: aggregate limits across a few dimensions that block an account entirely once a threshold is crossed, or nothing at all. Done properly, every quote is priced on its marginal contribution to the book’s risk, given everything already bound, across every dimension the system knows to look at, continuously.
Which brings me back to Machiavelli. The river of Fortune cannot be kept from flooding. What can be done is the work of fair weather: the dikes and canals that decide in advance where the water goes. That is the real purpose of insurance — to understand the sources of risk in the world, model them, and distribute them so they don’t fall too harshly on any one person or business.
An MGA or insurer distributing that risk has to take care not to fall victim to Fortune itself. As my cofounder Chris likes to say, anyone can make money in a good market; it’s when the tide turns that you see who has been underwriting with discipline. Technology and AI can certainly improve operational efficiency and metrics like time-to-quote, and we’re as excited about that as anyone. We intend to be best in class there. But we’re convinced that the real reinvention of this industry over the coming decade will come from learning from the best underwriters, combining and scaling that expertise through technology, and applying it not just to real-time risk selection but to real-time portfolio management — bringing specialty underwriting to a level of rigor that was previously impossible. At Vivere, we intend to lead that charge.