I spent twelve years as an actuary before I started Morildsen. The first six at a Scandinavian reinsurer, pricing catastrophe bonds and longevity risk. The second six as CPO at a Copenhagen digital insurer, where I watched the actuarial work I had done become the foundation of product decisions that were ultimately commercial rather than purely technical. When I moved to investing in 2019, I assumed the transition would be straightforward — the same mathematical intuitions, applied to a different asset class. I was partially right and meaningfully wrong in ways worth describing.
What Actuarial Training Gets Right About Venture Evaluation
The core actuarial skill is calibrating confidence to evidence. An actuary who is asked to price a new risk does not simply say "I don't have enough data" and stop — that is not a useful output. Instead, they build a framework: here is what can be inferred from analogous historical experience, here is the uncertainty range around that inference, here is how the credibility of the estimate improves with each additional data point, here is the minimum evidence threshold before the estimate becomes actionable. The output is always a distribution, never a point estimate presented with false precision.
This turns out to be highly applicable to seed-stage investment evaluation. When I look at a pre-revenue InsurTech with 18 months of operational history, I am not trying to produce a precise revenue forecast. I am trying to form a view on the shape of the distribution of outcomes — what does the 90th percentile look like if this works, what does the 10th percentile look like if it does not, and what is the range of middle scenarios that actually constitute the bulk of the probability mass. Founders who have only worked with venture investors sometimes find this framing unusual; most VC conversations focus on the upside scenario. Actuarial framing forces equal attention on the downside distribution.
The related skill is identifying what is observable versus what must be assumed. In catastrophe modelling, you separate the model components — hazard frequency, hazard intensity, vulnerability functions, exposure values — and you are very precise about which components are empirically calibrated and which are assumed. The compounding of assumption errors is a central risk in cat modelling; a small systematic bias in each of four components produces a dramatically wrong aggregate estimate. The same decomposition applies to a startup's unit economics model: which inputs are observed, which are analogies from comparable businesses, which are assumptions about future behaviour? Founders who cannot cleanly distinguish these categories are usually producing unit economics projections that feel precise but carry far more uncertainty than stated.
Where Actuarial Instincts Work Against You
The places where I had to consciously unlearn actuarial habits are equally important.
Mean reversion is usually correct in insurance, rarely correct in venture. Actuarial models assume that extreme outcomes revert toward historical averages over time — unusually bad loss years are followed by better ones, unusually profitable years attract new capital and competition that compresses margins. This is empirically reasonable in mature insurance markets. It is systematically wrong as a model for technology businesses, where network effects and compounding retention produce outcome distributions that are fundamentally non-mean-reverting. The best seed-stage companies I have backed did not revert to industry-average growth rates; they grew faster as they accumulated market position. Updating my prior on this took longer than I would like to admit.
Adverse selection intuitions can misfire. Actuaries are trained to be suspicious of any pricing or distribution innovation because novelty tends to attract adverse risk selection — the people who most eagerly adopt a new insurance product or channel are often those who know something about their risk that the new pricing model cannot yet see. This makes actuaries cautious about first-mover advantage claims from InsurTech founders. The caution is often warranted. But there are genuine cases where a new data signal or a new distribution channel permanently improves risk selection beyond what incumbents can achieve, and actuarial conservatism can cause you to discount those cases too heavily. I have had to develop a separate mental track for "adverse selection risk is real" versus "adverse selection risk explains why this specific team's innovation will fail to generate durable advantage."
The data sufficiency standard is too high. Actuarial sign-off on a new product typically requires several years of loss experience in reasonably similar segments before a pricing model is considered credible for deployment at scale. A seed-stage startup does not have several years of loss experience. Neither does a new product category. The investor's job is to evaluate whether the founding team has a plausible path to developing that experience base — whether they can build the loss data flywheel — rather than to withhold confidence until the data standard is met. Applying an actuarial sufficiency standard to seed investment decisions produces a portfolio of late-stage consensus bets, not seed-stage discovery bets.
The Judgment That Does Transfer
There is one actuarial capacity that transfers almost without modification to investment evaluation: the ability to hold a view on what a model is not capturing. Actuaries spend considerable effort on model limitations documentation — explicit statements about what the pricing model assumes, what assumptions are most sensitive to challenge, and what could cause the model to produce a materially wrong output if market conditions change.
When I evaluate an InsurTech founding team, I am implicitly running a version of this analysis on their product thesis. What does their business model assume about carrier behaviour, regulatory stability, data availability, or consumer adoption? Which of those assumptions is most vulnerable? Have they acknowledged the assumption explicitly and built contingency into their roadmap, or is it invisible to them? The teams I have the most confidence in are the ones who can articulate the failure modes of their own thesis clearly — not because they plan to fail, but because having thought clearly about failure modes is evidence that they have thought clearly about the business generally.
The honest version of this: the mathematical training is table stakes for evaluating InsurTech at technical depth. The useful part of the actuarial background is the epistemic discipline — the practice of being precise about what you know, what you are assuming, and what would change your estimate. That discipline is rarer in investment contexts than the mathematics, and it is the part I have not had to unlearn.