Morildsen Fund II closed in September 2023 at $63M. We want to share what we have updated in our thesis since Fund I — what the first portfolio taught us, how we see the European InsurTech landscape evolving through the end of the decade, and where Fund II capital will be concentrated.
Fund I deployed $32M across nine companies between 2020 and 2022. Three of those companies have since raised growth-stage capital. Two are approaching breakeven. Two we are still watching carefully. These are normal seed-stage distribution patterns. What the Fund I portfolio taught us is more interesting than the return profile at this stage.
What Fund I Taught Us About Thesis Fit
The Fund I thesis was broad enough to encompass InsurTech startups across distribution, underwriting, and claims — the full value chain. In practice, we found that our ability to add value, and therefore our ability to help companies compound, was substantially stronger in two areas: companies building AI-based underwriting and risk assessment models, and companies building the API infrastructure layer beneath digital insurance distribution.
In distribution-focused companies, the primary challenges are sales and go-to-market — building carrier partnerships, negotiating distribution agreements, growing an agent or broker network. These are important problems, but they are not where Astrid's technical depth or my actuarial background adds the most. The companies where we felt we had genuine differentiated value were those where understanding the model architecture, the data pipeline design, and the regulatory constraints on AI in insurance underwriting mattered directly to what the founding team was building.
Fund II concentrates this. We are writing fewer cheques but larger ones (€2.5M–€3.5M initial), and we are focusing on a tighter category definition: AI-native underwriting infrastructure and risk assessment platforms. Distribution-only InsurTech, digital MGAs without proprietary risk models, and pure claims automation without model-layer differentiation are outside our sweet spot for Fund II, though not out of the universe we track.
The AI Infrastructure Layer Is Now Investable at Seed
When we started Fund I in 2019, much of the AI infrastructure that InsurTech companies needed was custom-built because it had to be. Model training pipelines, feature stores, model monitoring infrastructure, explainability layers — these either did not exist as off-the-shelf components or existed only in forms designed for different industries. Insurance-specific model infrastructure had to be proprietary.
By 2023, the general-purpose ML infrastructure layer has matured substantially. A seed-stage InsurTech does not need to build its own model registry or its own feature store from scratch. What it needs to build — and what constitutes genuine competitive differentiation — is the insurance-domain data layer: the structured representation of policy, exposure, and claims data that makes ML features meaningful for insurance risk. The domain model, not the infrastructure beneath it, is now where proprietary value accumulates.
This changes the seed-stage investment calculus. We can now evaluate whether a founding team has a superior domain model from very early evidence — how they describe the features they are building, how they have structured the exposure data, what signals they are using that incumbents do not have access to. The build time to a demonstrable proof of concept has shortened. The window between founding and the point where we can make a substantive technical evaluation has compressed from 18-24 months to 9-12 months for the best teams.
European Geography: Why Now Is More Differentiated Than 2019
Our European focus was a deliberate choice in Fund I, not simply a function of geography. The thesis was that European insurance regulation — Solvency II, EIOPA oversight of AI in insurance, the distributed carrier landscape across 27 national regulatory regimes — created a substantially different competitive environment than the US market, and one that required different domain knowledge to navigate well.
Four years later, this thesis has strengthened rather than weakened. The EU AI Act's provisions for high-risk AI systems — which insurance underwriting models fall under — create a regulatory differentiation that US-originated InsurTech teams consistently underestimate. EIOPA's guidelines on the use of Big Data Analytics in insurance, first published in 2019 and updated since, impose explainability and fairness requirements that are substantively different from what US carriers operate under. Building a model that is both commercially viable and EIOPA-compliant requires genuine domain investment.
European founders who have built their models against these constraints from the start have a structural head start over US teams entering Europe. We see this acutely in commercial underwriting: a team that has spent two years building an AI underwriting model to EIOPA explainability standards is genuinely several years ahead of a US team attempting to adapt a US-compliant model for European deployment.
Where Fund II Capital Will Be Deployed
Three thesis areas concentrate the majority of Fund II capital:
Continuous underwriting infrastructure. The technical and regulatory conditions for underwriting policies that reprice continuously based on live data — rather than at annual renewal — are converging. The companies building the model infrastructure that makes this possible are at the most interesting seed stage right now. Annual policy structures are not a natural law; they are a technology limitation that AI-native infrastructure is beginning to remove.
Commercial lines risk APIs. Commercial insurance has lagged personal lines in API-accessible risk data by a meaningful gap. The platform companies building structured API access to commercial risk data — property exposure, fleet telematics, supply chain stress signals, business interruption triggers — are creating the data layer that the next generation of commercial underwriting models will run on. This infrastructure has natural monopolistic characteristics once established.
Claims model infrastructure for complex loss categories. Automated claims handling has progressed most in simple, high-volume, low-complexity categories — motor FNOL, straightforward property damage. The interesting territory for Fund II is ML-based assessment of more complex categories: liability claims, professional indemnity, business interruption, where the outcome is substantially affected by early triage quality and where the economic impact of better triage compounds significantly at portfolio scale.
The decade thesis has not changed: insurance is transitioning from pooled statistical loss pricing to continuous individual risk assessment. What has sharpened is our understanding of where in that transition the seed-stage infrastructure decisions are being made, and therefore where the foundational companies of the next phase are being built right now.