Winning in two realities: an AI strategy for insurers in an intermediary-led market

A perspective on where to play and how to win as agentic commerce reshapes insurance distribution, and why the obvious strategy can quietly become the risky one.

For many Dutch insurers, AI is no longer only an efficiency topic, but a commercial topic. Investments are growing, but only a small minority of players are turning them into measurable value. Our research into the Dutch market found that 41% of insurers and 29% of MGAs and brokers remain in an experimentation phase, with a further 12% only just beginning to build the foundations on which to apply AI [1]. The gap between leaders and the rest is widening.

At the same time, agentic commerce is becoming tangible, not theoretical. In the Netherlands, service provider Risk is publicly experimenting with AI-based comparison and purchase of insurance, part of a broader wave of Dutch intermediaries testing what agentic distribution actually looks like in practice [2]. Internationally, the examples are more numerous. Spanish digital insurer Tuio launched the first insurer-built AI insurance app on ChatGPT, delivering personalized home insurance quotes directly inside the conversation. U.S. comparison platform Insurify launched a ChatGPT-based motor insurance comparison app in parallel. On 9 February 2026, the market took notice: listed insurance brokers in the U.S. dropped on average 8.9% in a single day on concerns about AI-driven distribution, with European insurers following the next day [3].

Our own estimate is that by 2030, €4–8 billion in Dutch non-health insurance premium — roughly 10–20% of the market — will flow through genuinely new agent-driven channels: embedded platforms, autonomous comparison agents, and AI-mediated purchase journeys that bypass traditional distribution touchpoints entirely [4]. That is not a prediction of catastrophe for incumbents. But it is a meaningful shift in where premium originates - and for insurers whose products are not discoverable to those agents, it is invisible volume.

The boardroom question is the same whether you are a large carrier or a mid-sized broker:

What role do you play when AI agents, not humans, increasingly mediate advice and purchase?

"Our own estimate is that by 2030, €4–8 billion in Dutch non-health insurance premium — roughly 10–20% of the market — will flow through genuinely new agent-driven channels."

Where AI investment is flowing today

Before talking about strategic choices, it is worth looking at where insurers are actually putting their AI money.

Most investment goes into claims, underwriting and servicing. This is the operational core of the business, and it is a logical place to start: the data is rich, the ROI is measurable, and internal sponsors are easy to find. Our research shows this is where the bulk of AI value is being realized in the Dutch market today, and we would not argue with that prioritization. Claims and underwriting efficiency are necessary foundations, not distractions [5].

The front end of the business — advisor tooling, product discoverability, agent-accessibility — receives meaningfully less. This is a logical starting pattern, but it is starting to look like a blind spot. The advisor-facing and agent-facing capabilities that determine how an insurer shows up in an AI-mediated buying journey are the capabilities receiving the least investment at most insurers. In a market where the buying journey itself is being rebuilt, that asymmetry matters.

The leaders look different. A small group of AI transformers (fewer than one in five players in the Dutch market) combine clear C-level commitment, an AI ready operating model and data platform, reusable use cases, and measurable ROI. They are not necessarily spending more than their peers. They are spending across both the operational core and the distribution front end, and they are treating AI as a domain-level transformation rather than a collection of point solutions [6].

This direction isn’t wrong, but in our view it’s incomplete. And that incompleteness is exactly what an innovator’s dilemma looks like in practice.

The innovator’s dilemma, close to home

The familiar trap, that serving your best existing customers well is also what can leave you exposed, applies cleanly to incumbent insurers today.

Many of an insurer’s best customers are served through its intermediary channel. AI that makes claims faster or underwriting more accurate shows up immediately in loss ratios and broker satisfaction. AI that makes an insurer discoverable by ChatGPT, or that gives its products modular APIs for an agent to query, shows up nowhere measurable in the first two years. It has no natural sponsor. It may even look like it undermines the intermediary channel that pays the bills.

Run that logic forward and it’s easy to see how a well-run insurer arrives at 2028 having done everything its current customers and partners asked of it — and still find itself invisible to the agent-mediated channels that are starting to matter. This is not a hypothetical. It is the direction that current investments are pointing in.

The job is not to escape this tension (it can’t be escaped) but to name it, and make a deliberate choice about where to place your bets.

Who is building visibility into AI systems today, and who will soon be optimizing themselves to perfection… but be out of the picture?

The strategic choice: four postures

The real question for an intermediary-led insurer is not whether to invest in AI & agentic commerce, but how to split AI investment across two decisions that pull in different directions: how much goes into empowering the intermediary channel, and how much goes into building direct agentic channels. Those two axes define four distinct strategic postures.

Operator Play (bottom-left)

Compete on cost. AI goes mainly into operations: underwriting automation, claims straight-through processing, SG&A reduction. Distribution strategy and AI strategy stay separate. This is where a lot of Dutch insurers sit today by default: the current investment pattern, heavily weighted to the operational core, looks like an Operator Play even when no explicit choice has been made. It can work for a pure cost player with no differentiated advisory story, but as a standalone posture it assumes the channel economics stay stable, which is the one thing agentic commerce is unlikely to leave alone.

Partner Play (top-left)

Lean fully into the intermediary. Build the best AI-augmented tooling for brokers, advisors and tied agents in the market. Equip them with proprietary risk and portfolio data, embedded advisory logic, and autonomous service layers. Compete with the intermediary, not against them. The bet: advisory relationships and proprietary data compound value faster than agent-mediated disruption arrives. This is the posture most consistent with how most Dutch incumbents already win, and therefore the one most at risk of quietly optimizing itself into irrelevance. It also has a real execution dependency: the intermediary has to be able to absorb the investment. More on that below.

Direct Play (bottom-right)

Build for the agent economy. API-first products, modular coverage, machine-readable documentation, verifiable identity credentials, and embedded partnerships with non-insurance brands. Accept some channel conflict with intermediaries as the price of being visible when agents come shopping. The bet is that the window for establishing agent-visibility is short, and being late is hard to recover from. This is rarely the right only posture for an incumbent with significant intermediary revenue, but it is increasingly hard to avoid in some form.

Dual Play (top-right)

Run both, deliberately. Strengthen intermediary AI capabilities and build direct agentic infrastructure, with governance separation so the two don’t cannibalize each other. It is the most expensive posture and the most demanding to run, and the only one that hedges both futures.

The key point is that the two axes can be in real tension, and that tension is what the innovator’s dilemma quietly pushes insurers to mis-manage. A credible strategy names the tension and makes the allocation choice explicit.

Why intermediary maturity shapes which posture is actually available

The aggregate picture from our research is that insurers are somewhat ahead of brokers and MGAs on AI maturity, but on average does a lot of hiding. Three distinct groups sit inside the intermediary market, each with different implications for which posture is realistic.

Innovating intermediaries are experimenting actively with agentic comparison and API-based distribution. In some cases they are moving faster than the insurers they work with. Partner Play here is a genuine peer-to-peer collaboration.

Consolidating intermediaries are mid-way through consolidation, carrying IT complexity that is, as one insider described it to AMweb, “no longer explainable” [7]. Partner Play investment aimed at these organizations may land in an environment simply not ready to absorb it yet, no matter how good the tooling is.

Traditional intermediaries are still in early experimentation. Partner Play here means investing as much in change and process redesign as in the AI tooling itself.

For digitally mature insurers, the honest question is: how much further can you push Partner Play before the bottleneck moves from your side to theirs? At that point, the marginal euro is better spent on Dual Play or Direct Play; or on monetizing AI capabilities as a service to the broker market.

How the choice lands in the Dutch market

Three features of the Dutch market shape potential adoption of agentic commerce:

The intermediary channel is structurally important but unevenly ready. Most MGAs and brokers in our research are still early in their AI journey; a minority are moving fast. Insurers leading Partner Play need to match their investment profile to where each part of their network actually is, not where they wish it were [8].

Dutch consumers and SMEs are digitally fluent, and agent-mediated search will land quickly here. Comparison sites already handle a good portion of personal-lines distribution. Their evolution into agent-queryable services - or their displacement by consumer AI agents querying insurers directly - will happen faster in the Netherlands than in most European markets. The Direct Play exposure for Dutch insurers is higher than a continental average would suggest.

Regulation raises the floor and opens a first-mover advantage. eIDAS 2.0, the AI Act, and high standard for supervision simultaneously demand higher infrastructure and governance standards and open opportunities for insurers that build them properly. The compliance cost of doing it well is also the cost of being discoverable by enterprise AI agents operating under verified digital identity.

The practical reading for most Dutch incumbents is that pure Partner Play is not enough, even when it is the most natural position. The 12–18 month window for building agent-visibility is too short to address later, and the minimum-viable Direct Play hedge is cheap enough that skipping it is hard to justify.

So the question is narrower than the full matrix:

Partner Play with a Direct Play hedge, or Dual Play with explicit separation? The answer depends on your conviction about the pace of agentic disruption, your digital maturity relative to your intermediary network, and your organizational capacity to run two operating logics at once.

How to win, posture by posture

Each posture has its own winning condition and its own way of quietly failing.

How to win Partner Play

You win if your intermediaries become measurably more productive than your competitors’ intermediaries, and if that productivity is built on your proprietary data. The economic logic is the capacity dividend: agentic quote-and-bind, autonomous servicing and proactive renewal logic can together free up around 45% of a broker’s administrative operating base, which can then be redeployed into higher-value advice [9]. Two things go wrong in practice. First, the dividend gets competed away as price rather than reinvested in advice, unless the commercial model is designed to prevent it. Second, the intermediary must be able to adopt the tooling, so Partner Play leaders need to invest as much in broker enablement and change as in the AI itself. Where the intermediary is still integrating IT post-consolidation, that enablement starts with the backbone, not the AI.

How to win Direct Play

You win if any AI agent, consumer or enterprise, can discover, understand and transact with your products. The test is simple: when an agent searches, does your product show up in the consideration set? Insurers whose products and pricing aren’t queryable via APIs, whose documentation isn’t AI-comprehensible, and whose brand lacks verifiable credentials are structurally invisible, regardless of how good the product actually is. The required capabilities are infrastructural: API-first product architecture, modular coverage, structured machine-readable data, and verifiable identity. None of these are optional.

How to win Dual Play

You win on governance discipline. The two operating models are genuinely different — different metrics, different margins, different partners, and different customer acquisition logic. The most common failure mode is a Dual Play in name only, where the agentic track is starved by the day-to-day economics of the intermediary channel. A credible Dual Play needs a separate P&L for the agentic track, separate leadership with a clear mandate, and a shared data foundation that both tracks can draw from without the intermediary channel holding a veto on data access.

How to win Operator Play

You win on ruthless cost discipline combined with a realistic plan to exit advisory-led lines. This is a deliberate bet, not a default — and most insurers who end up here ended up here by not choosing, which is a different thing entirely.

Four capabilities that matter in any posture

Posture determines where you place the bet. Execution determines whether the bet pays off. The evidence is consistent that what separates AI winners from the rest is rarely the technology itself. The Stanford Digital Economy Lab’s study of 51 enterprise AI deployments found that the real differentiators are change management, adoption, and data quality, the unglamorous line items most organizations under-invest in precisely because they are not about the technology [10]. Four capabilities translate that into practical reality for insurers.

Proprietary data on a unified platform. Most insurers don’t have a single view of customer, risk and intermediary performance. Without it, Partner Play advice is generic, Direct Play agent-accessibility is thin, and Dual Play can’t reuse data across the two tracks. The data platform is the gating capability, not one of several.

API-readiness and product modularity. For Partner Play, APIs are what make broker tooling fast and autonomous servicing real. For Direct Play, they are the product. For Dual Play, they are the shared infrastructure. Insurers without API-accessible products will lose ground in both intermediary and direct channels as broker tooling and customer agents alike start requiring machine-readable data.

Change capacity on both sides of the relationship. Advisor tooling creates no value if advisors don’t adopt it, and agent-accessible infrastructure creates no value if the organization can’t sustain the discipline of keeping it machine-readable. For Partner Play specifically, the adoption challenge extends beyond the insurer’s walls into the intermediary network — and that is where a lot of Partner Play investment quietly underperforms.

Management systems that match the ambition. AI-enabled distribution needs AI-enabled steering. Real-time performance monitoring, traceability of AI decisions, and exception-based management are the minimum conditions under which leadership can trust AI-augmented processes [11]. Without them, human oversight bottlenecks quietly strip the economics from the AI investment.

What to do in the next twelve months

For an insurer in the position most Dutch incumbents occupy — strong intermediary relationships, AI investment concentrated in the operational core, limited direct agentic exposure — four priorities stand out.

1) Make the posture choice explicitly. Name the innovator’s dilemma rather than navigating around it. An unmade choice becomes Operator Play by default, which is the worst of the four — and it is where the current investment pattern is already taking most insurers.

2) Segment the book honestly. Different lines of business face different disruption profiles and warrant different postures. Personal lines and standardized commercial are more exposed to Direct Play displacement than complex commercial or D&A. The same insurer can legitimately run Partner Play on D&A and Direct Play on personal auto.

3) Fund the minimum-viable Direct Play hedge now, regardless of core posture. API-first product exposure, structured machine-readable data, and discoverability testing against consumer AI agents are not strategic bets. They are the cost of staying in the consideration set.

4) Match the Partner Play profile to real intermediary capacity. Before scaling advisor tooling, know which of your partners is innovating, which is stuck on post-consolidation integration, and which is still in early experimentation. Design the investment profile (and the change investment around it) accordingly. The commercial model also matters: design it so freed broker capacity flows into advisory value the customer will pay for, not into price competition.

The honest conclusion

The future of insurance distribution won’t be defined by a single model. The insurers who win won’t be the ones who picked the “right” posture — they’ll be the ones who picked a posture deliberately, funded it properly, and resisted the pull of their most successful current customers long enough to invest in the ones they don’t yet have.

Today’s position (AI concentrated in the operational core, underfunded at the distribution front end, matched unevenly to intermediary readiness) isn’t a strategy. It’s the absence of one. The innovator’s dilemma doesn’t resolve itself quietly. It gets worse the longer it’s unmade as a choice. The frameworks exist. The technology is proven. What’s left is the decision.

Author: Casper van Hilten

Curious about the next steps for your AI strategy? Let's talk.

Jan-Pieter van der Helm

Managing Director Insurance

janpieter.vanderhelm@igh.com

+31622554190

Sources:

[1] IG&H, The state of AI in the Dutch insurance market: the road to returns (December 2025). Sector-wide research across 8 insurers, 9 MGAs/brokers and 5 insurtechs.

[2] AMweb, Verzekeringen vergelijken en afsluiten via AI: sector experimenteert driftig (2026), on Dutch intermediary experimentation with agentic commerce, including Risk.

[3] MarshBerry, Is AI rewriting the rules of insurance distribution and valuation? (February 2026), on the 9 February 2026 broker-stock repricing and the Tuio and Insurify ChatGPT app launches. See also Torrance, S., Agentic Commerce: The Insurance Distribution Takeover, ai-risk.co (October 2025), and IG&H, AI Impact on Insurance Distribution (February 2026).

[4] IG&H scenario estimate, based on Verbond van Verzekeraars market data and analysis of emerging agentic distribution patterns.

[5] Ibid.

[6] Ibid.

[7] AMweb, Consolidatie leidt tot veel te complex IT-landschap. ‘Niet meer uit te leggen’ (2025), on IT integration challenges for consolidating Dutch intermediaries.

[8] Ibid.

[9] IG&H project experience, illustrative capacity-reallocation model for mid-market brokers: agentic distribution, service automation and autonomous renewals together reduce the administrative operating base by approximately 45%, with the opportunity to reinvest that capacity into advisory work.

[10] Pereira, E., Graylin, A.W., Brynjolfsson, E., The Enterprise AI Playbook: Lessons from 51 Successful Developments, Stanford Digital Economy Lab (April 2026).

[11] Martin, R., Becoming an AI Augmented Enterprise: How to Leverage AI Strategy (March 2026), rogerlmartin.substack.com, on AI as a functional strategy and the role of enabling management systems. See also Brynjolfsson, E., The Turing Trap: The Promise & Peril of Human-Like Artificial Intelligence, Daedalus (2022), on the economic case for augmentation over substitution. Further market context: BCG, Competing for the AI-Empowered Insurance Customer (2026) and AI-First Companies Win the Future: Life Insurance & Annuity (January 2026).