Most conversations about insurance and startups begin with the wrong question. They ask whether the group should have a venture arm. That question has no useful answer, because there is no single thing called a venture arm. There are four distinct structures, they solve different problems, and they cost very different amounts of governance to run.
An insurer that picks the wrong one does not usually discover the mistake in the portfolio. It discovers it in the friction: a committee that cannot approve anything on a startup's timescale, or a business unit that never adopts what the venture team bought. This article sets out the four models, what each one is actually for, and what each one costs before any capital is deployed.
Model One: The Venture Client, Buying Before Investing
The first model involves no equity at all. The insurer becomes a startup's customer, pays for a pilot, and measures whether the product solves a real internal problem.
The clearest public example in insurance is the Zurich Innovation Championship. Zurich describes the programme, which has run since 2018 and reports more than 150 projects across more than 30 countries, in terms that leave no ambiguity about the mechanism: "Unlike traditional Venture Capital models, we invest in the adoption and validation of a startup's product, service, or technology, without taking equity." The funding it offers is explicitly equity-free.
What this model buys is speed and evidence. A pilot can be authorised by a business unit budget holder rather than an investment committee, and the outcome is a fact about whether the product works inside the organisation rather than a valuation opinion.
What it does not buy is any claim on the upside. If the startup becomes a category leader, the insurer has a supplier relationship and nothing else. That is a deliberate trade, and it is the right one when the strategic question is "does this work for us" rather than "do we want to own part of this".
Model Two: The Dedicated Investment Unit
The second model is the one most people picture: a distinct entity, its own team, its own mandate, investing the group's balance sheet.
Allianz X is a documented example at scale. It presents itself as the strategic investment arm of the Allianz Group and reports 25 current direct investments, more than 1.7 billion in assets under management, five full acquisitions, more than ten exits, and 40 employees across five continents.
Two things are worth noticing in that description. The first is the size of the team relative to the number of positions. A dedicated unit is not a light structure, and the staffing is the real cost, not the capital. The second is that the stated purpose runs beyond returns: the unit exists to build connections between portfolio companies and the group. That is the defining feature of this model, and also its main point of failure. A dedicated unit that produces good returns but no adoption inside the group will eventually be asked what it is for.
Model Three: The External Mandate
The third model delegates the investment function to an external manager operating under a mandate written by the insurer. The insurer keeps the thesis, the strategic direction and usually the co-investment rights. It does not build the team.
This model is deliberately harder to document publicly than the first two, and the honest position is that no insurer mandate can be cited here with a public source. These arrangements are commercial contracts between two private parties, and neither side has an incentive to publicise the terms. Their existence is not in doubt; their details are simply not public.
What can be said structurally is what the mandate has to settle, because that is where the model succeeds or fails: the boundaries of the thesis, who can say yes and how fast, how fees relate to strategic outcomes rather than deployment volume, and what happens at the end of the term. All of that belongs in writing before the first deal, not after the first disagreement, and it is worth reading alongside what the mandate should specify in detail.
The trade in this model is control against speed and cost. The insurer gives up direct authorship of each decision and gains a functioning team without a multi-year hiring cycle. Whether that is a good trade depends almost entirely on how many deals per year the group realistically expects, which is the same variable that drives the cost of each approach.
Model Four: Investing Through Third-Party Funds
The fourth model puts the insurer in the position of investor in someone else's fund, rather than direct investor in companies.
This is the most common and least discussed of the four, because it looks like an allocation decision rather than an innovation decision. It offers diversification, no operational burden, and access to deal flow the insurer would never see alone. It offers almost nothing in strategic proximity: a position in a fund does not give a business unit a reason to talk to a portfolio company, and the reporting arrives quarterly, long after the moment when a commercial conversation would have been useful.
This model is best understood as a complement rather than a competitor to the other three. It is a reasonable way to learn a sector before committing to a structure, and a poor substitute for one.
What Each Model Really Costs in Governance
Capital is rarely the binding constraint for an insurer. Governance is.
The venture client model is the cheapest to govern because the decision sits with whoever owns the budget for the pilot. The dedicated unit is the most expensive, because it requires a permanent team, an investment process, a committee, and a reporting line that has to survive changes in group leadership. The external mandate moves most of that cost outside, but replaces it with the harder task of writing a mandate precise enough to be enforceable. Investing through third-party funds is the lightest of all, and correspondingly the least strategic.
The pattern worth remembering: the more strategic proximity a model offers, the more internal governance it consumes. There is no structure that delivers both close alignment and low overhead.
How to Choose Without Guessing
Three questions settle most of the decision.
What is the real objective? If the group needs to know whether a technology works in its own operations, the venture client model answers that faster and cheaper than any equity structure. If the objective is a claim on the upside of a sector, it does not answer it at all.
How many decisions per year? A group that expects two or three investments a year cannot justify a permanent team, and should be looking at a mandate or at fund positions. A group expecting a genuine flow of transactions can amortise a dedicated unit.
Who inside the organisation will actually adopt the outcome? This is the question most often skipped, and the one that decides whether the programme survives its third year. A model that produces investments nobody internally uses will be defended by its team and by nobody else.
Frequently Asked Questions
Can an insurer run more than one model at once?
Yes, and many do. The combination that works is a venture client programme feeding evidence into an equity structure, so that investment decisions are informed by whether the product actually performed inside the group. The combination that fails is two equity structures with overlapping mandates competing for the same deals.
Is the venture client model just a pilot programme with a new name?
The mechanism is similar. The difference is intent and follow-through: a venture client programme is designed so that a successful pilot leads to a commercial contract, rather than ending in a report.
Does a dedicated unit need to sit outside the insurer legally?
Not necessarily, but it needs to sit outside the insurer operationally. The recurring failure is a unit that reports into a structure whose approval cycle is slower than the market it is investing in.
What is the fastest model to start?
The venture client model, by a wide margin. It requires no vehicle, no committee and no capital allocation, which is also why it is the most common entry point for insurers testing whether they want a venture strategy at all.
Conclusion
The four models are not stages on a maturity path, and an insurer does not graduate from one to the next. They are answers to different questions: whether a technology works, whether to own part of a company, whether to delegate the function, or whether to buy exposure to a sector.
The decision that matters is not which model is best in the abstract. It is which question the group is actually trying to answer this year, and whether it has the governance to support the answer it chooses. Getting that pairing wrong is what produces venture programmes that look active for two years and quietly disappear in the third, a pattern visible in how the first three years of a programme tend to unfold.