Before I became an investor, I spent twelve years pricing things that might not happen. Catastrophe bonds, excess-of-loss reinsurance treaties, aggregate stop-loss contracts on commercial fleets. The work is methodologically humbling in a specific way: you are always reasoning about distributions, not point estimates. You are always managing the gap between what the model says and what the data will eventually reveal.
That experience changes how you evaluate a seed-stage insurtech. In ways I did not fully anticipate when I founded Morildsen.
The Distribution Intuition
A standard venture evaluation asks: is this market large? Is this team capable? Is the product differentiated? These are necessary questions. But for insurance-adjacent technology, there is a prior question: does this founding team understand that insurance deals in distributions, not expectations?
Virtually every insurance technology product that fails in production fails for the same reason. The team built for the expected case. They optimized for median performance — mean claim severity, average fraud rate, typical policy tenure. The actuarial practitioner knows that insurance businesses are destroyed by the tail, not the median. A flood year with 3-sigma rainfall. A liability claim that was classified as routine at FNOL and turned into a £4.2M reserve. A fraud ring that learned to pattern-match your detection thresholds.
When I meet a founding team, I listen for whether they are reasoning about distributions or means. It is a quiet signal but it is rarely wrong.
Model Validation vs. Model Deployment
Twelve years of pricing catastrophe risk taught me the difference between a model that performs well on historical data and a model that behaves correctly in production. In catastrophe modelling, the canonical failure mode is a model that has been calibrated on the period 1990–2010 and encounters a loss pattern outside that regime. The model is technically correct on its training window. It is functionally useless for the event you need it for.
Insurance ML startups reproduce this failure at speed. They train on whatever carrier data they can access, achieve impressive AUC scores, and present that to carriers as proof of deployment readiness. The carrier signs a pilot. Six months later, the model's performance degrades because the data generating process has shifted — telematics hardware changed, customer acquisition channel changed, macroeconomic conditions moved claim frequency. The founding team has no machinery to detect this. They do not have actuary-grade monitoring protocols.
This is not an indictment of the teams. It is a design gap. ML engineers and actuaries are not the same profession, and most insurtech founding teams are heavily weighted toward the former. What I look for at seed is not perfection — it is whether the team has the intellectual framework to know what they do not yet know.
The Pricing of Optionality
The actuarial background gives something else: comfort with pricing optionality under uncertainty. This sounds abstract. It is practically very useful.
When we are evaluating a seed investment, we are not buying a known cash flow stream. We are buying a set of options — the option to participate in Series A, to support a pivot, to observe adjacent markets that did not exist at the time of investment. The frameworks actuaries use for pricing embedded options in insurance contracts (surrender options, extension options, reinstatement provisions in reinsurance) translate surprisingly directly to the option-pricing logic of seed investing.
Specifically: we are never paying for the expected outcome of the current product. We are paying for the distribution of outcomes across all plausible futures of this team operating in this space. The math is not precise — the inputs are far too uncertain for that. But the mental model is correct. And I think it makes me a more patient investor than some peers, because I am not anchored to a single trajectory.
Where the Actuarial Mindset Has to Be Unlearned
This is worth saying clearly, because the actuarial frame is not always an advantage.
Actuaries are trained to be conservative. We are paid to not be wrong on the downside. The profession's entire incentive structure rewards underestimating exposure rather than overestimating opportunity. The bias is structural. Every pricing model I built in my reinsurance years was built to hold reserve adequacy, not to capture upside.
Venture capital requires the exact opposite. The return distribution is so skewed that excessive caution about downside — declining the investment in a team that turns out to be an outlier — is the most expensive error you can make. We are not saying the actuarial habit of extreme caution is wrong in insurance contexts. We are saying it is genuinely harmful if applied unchanged to seed-stage investing decisions.
I learned this through the first fund. We passed on three investments in 2020 that we had scored as "technically sound but founder maturity concerns." Two of those companies went on to raise Series B rounds at valuations that would have represented 18–22x on our seed entry. The caution was real. The cost was real. We recalibrated.
What This Means for How We Evaluate Teams
The mathematical heritage gives us a specific lens when we sit across from a founding team: we are not evaluating whether they have the right answer. We are evaluating whether they have the right epistemic relationship with uncertainty. Do they know where their model is confident and where it is extrapolating? Do they have a theory about what will break, or only about what will work? Can they tell us where the data is thick and where it is thin?
A team that can answer those questions at seed stage — before they have production data, before they have their first carrier partnership, before they know what FNOL-to-payment conversion rate they can actually achieve — that team has the intellectual architecture to build something durable in insurance.
That is what the actuarial background lets me evaluate. Not the model. The thinking behind the model.
The best seed rounds we have done at Morildsen share a pattern: the founders understood what they did not know before we pointed it out. That is not a common thing. The mathematical heritage helps us find it.