We have now made 14 investments across two funds, spanning seed and pre-seed rounds in European InsurTech from 2020 to 2026. Looking back across the portfolio with enough time to see meaningful divergence in trajectories, three patterns distinguish the rounds that generated the most durable companies. They are not the patterns that dominate investor conversation — they are less legible at the time of investment and more visible in hindsight. Writing them down feels useful, both as a record of what we have learned and as a signal to founders about what we are actually looking for.
Pattern One: The Team Had Already Built the Wrong Version
The strongest founding teams we have backed had, almost without exception, already built something that did not work before they showed up to pitch the thing we invested in. Not a catastrophic failure — more typically, a product that revealed a fundamental assumption error about the market they were entering, which forced them to reframe the problem before rebuilding.
In practical terms: a team that spent six months building a direct-to-consumer motor insurance product, acquired 800 customers, and discovered that their loss ratio was structurally higher than their pricing model assumed — because the customers who adopted digital-first motor insurance skewed toward a risk profile they had not adequately modelled — has learned something irreplaceable about insurance distribution and adverse selection. When they pivot to building carrier-facing risk assessment infrastructure, they bring an understanding of why the insurance product economics failed that most software engineers without insurance background do not have.
We are not saying failure is a prerequisite. We are saying that the founders who arrived with their first and only idea perfectly formed and polished were rarely the ones who built the most durable companies. The learning from an early wrong version tends to show up in product architecture choices, in the nuance of how founders describe the problem they are solving, and in the speed with which they identify when a new assumption is being proved wrong.
Pattern Two: Carrier Relationship Quality Was Asymmetric to Deal Stage
Across the portfolio, the companies that subsequently attracted the most durable commercial traction shared a characteristic at seed stage that was not always obviously legible: they had unusually deep relationships with one or two incumbent carriers, disproportionate to their company stage and size.
Not "carrier X has agreed to pilot" — many early-stage InsurTechs have that. We mean carriers where a Chief Underwriting Officer or Head of Claims had personally championed the engagement, had made internal budget commitments, and had structured the pilot in a way that was oriented toward production deployment rather than exploratory proof-of-concept. The distinction is visible in the nature of the data sharing agreement, in who on the carrier side is attending meetings, and in whether the pilot has defined success metrics tied to production deployment criteria.
This pattern predicts durable commercial traction better than seed-stage revenue. A pilot with genuine internal sponsorship at a tier-1 carrier is worth more than €200K in early-adopter revenue from smaller carriers who are piloting everything. The latter generates ARR. The former generates the production deployment reference that catalyses the next three carrier conversations.
Pattern Three: The Technical Co-Founder Understood Model Limitations Better Than Model Capabilities
The strongest technical founders we have met in InsurTech can articulate, clearly and without prompting, the specific conditions under which their model will perform worse than they want it to. Not in a defensive, hand-waving way — in a precise, "here is the distributional shift that will cause accuracy to degrade and here is how we will detect it" way.
This matters because insurance is a domain where model limitations have financial and regulatory consequences. A fraud detection model that becomes less accurate as fraudsters adapt to its signals is not an abstract problem — it is a claims loss ratio problem that affects the carrier's combined ratio and ultimately the risk of losing the deployment contract. A technical founder who cannot describe the adversarial dynamics of their own model, or who is confident that their model will maintain accuracy as the underlying distribution shifts, has either not thought carefully about the problem or is not being honest about it.
The founders who describe their model limitations clearly are not being modest about their technology. They are demonstrating domain depth — specifically, they understand that insurance risk models operate in non-stationary environments where adversarial adaptation, macro risk shifts, and regulatory changes continuously alter the problem the model is solving. That understanding is embedded in the architecture decisions they make: how they design data pipelines to detect distribution shift, how they build human-in-the-loop validation for edge cases, how they structure the feedback loop between claims outcomes and model retraining.
What These Patterns Have in Common
The thread that connects the three patterns is a specific kind of epistemic honesty about what the founder does and does not know. The founder who has already built the wrong version knows something about what they assumed incorrectly. The founder with deep carrier relationships knows something about what production deployment actually requires. The technical founder who understands their model's limitations knows something about what will break in production.
This is the actuarial sensibility applied to founding team evaluation: I am not looking for founders who have all the answers. I am looking for founders who have accurately identified what they do not yet know, and who have a credible path to resolving those unknowns before they matter. Founders who present as certainty what is actually assumption are harder to back, because they are less likely to make the adaptations their companies will inevitably require.
None of this is observable from a pitch deck. It emerges from conversations where we go deep into the specific technical and commercial decisions that have already been made — not to interrogate, but to understand how the founders think when the answer is not clean.