From super apps to autonomous ecosystems: Why AI is rewriting the operating and business model of insurance

This article is co-authored by Illia Pinchuk, CEO of DICEUS, and Dr. Evangelos Avramakis, an expert in strategic foresight, innovation, and AI.

Combining practical experience in insurance technology with expertise in strategic foresight and business transformation, the authors explore how AI is reshaping insurance operating models, moving the industry from digital service aggregation toward intelligent ecosystems built around customer outcomes, risk prevention, and continuous learning.

For insurance leaders, the AI debate is quickly moving beyond productivity. The more consequential question is what happens when AI does not simply support a transaction, but increasingly helps decide what should happen next – for a customer, a business or a connected asset. That shift changes who coordinates the customer journey, where risk decisions are made and, ultimately, where value sits. For insurers, the opportunity therefore goes beyond doing today’s work faster. It is about determining what role insurance should play when intelligent systems increasingly connect decisions, services and risks around the outcomes customers actually care about.

The operating assumption that AI is breaking

For much of the past decade, insurers viewed ecosystems primarily as a distribution opportunity. The rise of WeChat, Alipay, Grab, Gojek and Revolut reinforced a simple idea: whoever controlled a high-frequency digital interface could also control access to customers. Insurance therefore moved into travel bookings, e-commerce, banking and mobility journeys. The strategic question was largely about presence: how could an insurer become part of a journey that somebody else had already assembled? That response made sense in the platform era, when the prize was lower acquisition friction, greater transaction volume and access to customers at the moment protection was relevant. Grab still illustrates this model clearly: it combines deliveries, mobility, payments, lending and insurance in one application across more than 900 cities in eight Southeast Asian countries.[2]

The next technological wave is different because it begins to change the coordination itself. Generative AI can interpret unstructured information; agentic AI can plan and execute multistep work; connected assets generate continuous telemetry; and digital twins can simulate how a person, vehicle, building or process may behave as conditions change. Together, these capabilities do more than improve productivity inside an insurer. They make it easier to connect decisions across systems, companies and physical assets – and to react as circumstances change. McKinsey describes the progression from AI tools that augment tasks to end-to-end workflow automation and AI-first agentic systems.[3] For business leaders, the implication is important: activities that were historically separated because coordination was slow, manual or expensive can increasingly be brought together around a common objective.

This is the central insight from Sangeet Paul Choudary’s recent strategy argument. Businesses may have treated modularity as a permanent law when, in many settings, it was a response to high coordination costs. As intelligence inside individual modules becomes abundant, advantage migrates toward firms that connect activities, learn across boundaries and improve the whole system. The new form of integration does not require ownership of every asset. It requires ownership of the feedback loops that link sensing, decision, action and learning.[1]

For insurance leaders, this changes the starting point. The question is no longer only how AI can make underwriting, claims or service more efficient. A more fundamental question is emerging: if customers increasingly organize their decisions around outcomes such as reliable mobility, better health, secure retirement or resilient homes – and if AI increasingly helps coordinate how those outcomes are achieved – what role should insurance play in that system? The policy remains essential. But it may increasingly sit within a broader combination of risk intelligence, prevention, protection, assistance and recovery. This is the shift explored in the following pages: from insurance as a product inserted into a transaction toward insurance as an active risk capability within a continuously coordinated ecosystem.

AI changes the economics of coordination

Traditional business architectures were built around a practical constraint: coordinating many independent actors was difficult and costly. A patient had to connect doctors, laboratories, pharmacies, insurers and wellness providers. A traveler had to combine airlines, trains, local transport, payments and disruption cover. A homeowner had to coordinate lenders, utilities, maintenance providers, security systems and insurers. Each handoff introduced search costs, incompatible data, contractual friction and delayed decisions. Firms therefore optimized their own modules and relied on relatively stable interfaces between them.

Agentic AI weakens that constraint. An agent can translate between documents and data schemas, interpret policies and preferences, invoke tools and APIs, monitor changing circumstances and escalate exceptions to humans. Coordination may begin before every participant agrees on one common standard. This is strategically important because an AI-enabled translation layer can start as an interoperability service and later become the place where decisions are made. Once it controls the decision point, it can influence the rules, interfaces and economics of the surrounding system.[1]

Imagine a business traveler who says, “Get me to Warsaw by 9 a.m., keep the total cost below CHF 1,500, minimize disruption and ensure I am protected if the journey fails.” A conventional app can display flights, trains, taxis and insurance options. An agentic system can interpret the objective, assemble a multimodal journey, check corporate policies, assess weather and infrastructure risks, purchase appropriate protection, monitor the trip and rebook when conditions change. The customer no longer coordinates the sequence. The system coordinates toward an outcome.

This does not mean that broad autonomous ecosystems are already mature. Current deployments remain concentrated in bounded workflows, and recent governance research emphasizes the need for classification, evaluation, continuous monitoring and proportionate safeguards as agent autonomy increases.[8] The direction of travel nevertheless matters. When coordination becomes cheaper, industry boundaries become more permeable, and the strategic center moves from the provider of a component to the orchestrator of the outcome.

Super apps aggregate services; ecosystems orchestrate outcomes

The terms “super app” and “ecosystem” are often used interchangeably, but doing so obscures the real architectural shift. A super app is a distribution and engagement architecture. It places many services behind one identity, wallet and interface, reducing the need to switch applications. Academic research describes super apps as multi-purpose transaction platforms that connect multiple user groups and services in one environment.[4] An outcome ecosystem is broader. It coordinates independent capabilities, physical assets, data and incentives around a domain-level result such as mobility, health, housing resilience or retirement security. A super app may participate in such an ecosystem, and may even become an orchestrator, but the concepts are not synonymous.

Dimension	Super app	AI-enabled outcome ecosystem
Unit of value	A transaction or service accessed in one application	An achieved outcome, such as safe mobility or improved health
Center of gravity	Interface, identity, wallet and customer traffic	Intent, context, orchestration and feedback loops
Customer role	Navigate, compare and select services	Specify goals, constraints and permissions
Coordination model	Predefined journeys, mini-apps and platform rules	Dynamic planning across providers, agents and physical assets
Insurance role	Embedded add-on at a transaction point	Continuous risk capability inside the outcome
Examples	Grab, Gojek, WeChat, Alipay	Mobility, health, housing and retirement ecosystems

The difference can be seen in health. A health-oriented super app can make appointments, telemedicine and medication purchasing more convenient. A health ecosystem seeks to improve a person’s longitudinal health. It connects diagnostics, providers, wearables, behavioral interventions and financing around changing health needs. Discovery’s Vitality model offers an early, imperfect but instructive example of this logic: the insurer uses behavioral science, incentives and data to influence health, driving and financial behaviors rather than waiting passively for claims. An independent RAND Europe study covering more than 400,000 people found that the combination of Vitality incentives and Apple Watch was associated with a sustained 34 percent increase in activity.[5] The case is not evidence that every ecosystem intervention works, but it demonstrates how an insurer can move upstream from reimbursement toward measurable behavior change.

The same distinction applies to mobility. Grab organizes rides, deliveries and financial services in a common app.[2] A mobility ecosystem would optimize whether a person or shipment reaches its destination safely, affordably and reliably across public transport, private vehicles, charging networks, infrastructure and insurers. Tesla’s Safety Score illustrates one component of such a feedback loop: vehicle-generated measures are used to estimate future collision likelihood, provide drivers with feedback and potentially influence insurance cost.[6] The strategic advantage is not merely the score or the policy. It is the learning system connecting vehicle use, risk measurement, driver behavior and future decisions.

Autonomous ecosystems move the unit of value from products to outcomes

Incumbents typically respond to a new technology by adding it to the current product. An insurer may add a conversational assistant to a policy journey, automate a claims document or produce a faster quote. These improvements are useful, but they preserve the existing unit of value. AI-native competition starts with a more disruptive question: does AI allow the offering to be organized around a different object? Sangeet’s example of Figma is revealing. The strategic difference was not simply a better interface; design elements became reusable organizational objects inside a collaborative system rather than content trapped in isolated files.[1]

Insurance faces an analogous choice. The policy has historically been the organizing object because coverage, pricing, administration and regulation were built around discrete risk contracts. In a connected ecosystem, the more meaningful object may be the journey, the health trajectory, the home’s resilience state or the retirement income outcome. Protection can then be assembled from modular coverages, prevention services, guarantees and assistance capabilities that adapt as context changes. The policy does not disappear, and regulatory obligations remain. But it becomes one component of a wider outcome architecture.

Case study: From fixed products to context-driven protection

The insurer faced a practical product-management problem: more personalized offers meant creating and maintaining more product variations, while the existing core platform was not designed to rebuild a proposition every time a customer’s context changed. Keeping a separate configuration for each combination of coverage, limits, riders and pricing rules would have made the product catalogue increasingly difficult to manage. Replacing the core was not a realistic option either.

DICEUS addressed this by separating the configurable product logic from the underlying core. Coverages, limits, riders and pricing rules were modeled as individual components in a modular insurance layer connected to the existing environment. An AI-assisted layer could then use customer and risk signals already available in the journey — such as a new asset purchase, a life event, claims history or a change in usage – to identify relevant components and prepare a specific proposition. The core continued to handle the underlying insurance processes, while the modular layer determined how approved product components could be assembled for the situation at hand.

In practice, this changed how a new customer need could be handled. If an existing customer acquired a new asset or their circumstances indicated a relevant additional protection need, the insurer no longer had to start from a separate pre-built product configuration. The system could surface a proposition assembled from already approved components, while cases requiring additional judgment were sent to the broker or carrier for review. The broker’s role therefore shifted from manually originating every proposition toward reviewing, approving and contextualizing the recommendation before it reached the customer.

The result was a reusable product architecture rather than a growing catalogue of fixed variations. The same approved building blocks could be recombined across different customer contexts, the existing core remained in place, and the insurer retained control over pricing logic, exceptions and final decisions. This gave the business a practical way to personalize propositions without hard-coding a new product variation for every scenario or replacing the systems already running core insurance operations.

Digital twins strengthen this shift by allowing ecosystems to represent and simulate evolving real-world conditions. A twin combines data about an object, person, process or system with models of how it behaves. In insurance, potential applications include running what-if scenarios, predicting failures, generating synthetic data and testing interventions. The Actuary notes that digital twins can combine IoT, GPS, satellite and other data with historical information to improve prediction when the past is an unreliable guide.[7] Deloitte similarly describes twins as increasingly useful in dynamic environments, from manufacturing to medicine.[9]

The practical implication is a move from retrospective classification to prospective navigation. A mobility insurer can use vehicle condition, route, weather and driver behavior to recommend a safer action before a trip. A property insurer can combine sensors, maintenance records and weather forecasts to trigger a preventive intervention. A health insurer can support adherence or recommend a care pathway before deterioration becomes hospitalization. The strategic value lies not only in predicting risk, but in connecting prediction to timely action and measuring whether the action changed the outcome.

This also changes the boundaries of the insurance business model. Revenue may come from a combination of risk premiums, orchestration fees, prevention subscriptions, outcome-based contracts, data-enabled services and assurance for autonomous systems. The shift should not be romanticized: continuous data can create privacy, discrimination and consent risks, and not every risk is observable or preventable. But it opens a wider design space than simply selling a static product through a new channel.

Competitive advantage moves when scarcity moves

Strategy follows scarcity. A capability creates rents only while it is difficult to access or reproduce. AI makes many forms of competent output cheaper and more plentiful: product comparisons, content generation, routine analysis, first-level service, claims document review and basic coding. An insurer that celebrates faster production without asking whether that production is becoming economically abundant may optimize an activity whose strategic value is declining. Operational excellence can become strategically dangerous when it accelerates success in an old game.[1]

As intelligence becomes abundant, four constraints become more important. The first is context: high-fidelity, permissioned information about a customer, asset and operating environment. The second is orchestration: the ability to connect capabilities and adapt decisions across organizational boundaries. The third is trust and accountability: evidence that an agent acts within authority, explains its limits and can be audited. The fourth is risk-bearing capacity: the regulated capital and claims capabilities required when prevention fails. WEF research on AI agents emphasizes that autonomy, system integration and trust create governance demands that go beyond model performance, while its later work argues that durable trust depends on legible reasoning and bounded agency.[8][10]

A fifth source of advantage cuts across all four: the measurement and feedback architecture. Sangeet describes this as “the machine that builds the machine.” When outputs commoditize, advantage moves to the system that observes outcomes, learns what caused them and improves production faster than rivals.[1] For insurers, this means the valuable data is not simply a larger customer file. It is causal and operational evidence: which intervention reduced loss, under what conditions, for which segment, and with what unintended effects.

Scarcity shifts from…	Toward…	Insurance implication
Information and basic analysis	Permissioned context and causal insight	Build data rights, telemetry access and intervention evidence
Product availability	Outcome orchestration	Design modular protection and prevention around journeys or life goals
Customer access through channels	Algorithmic eligibility and ecosystem relevance	Become discoverable, explainable and reliable to human and machine buyers
Static expertise	Learning speed across boundaries	Own feedback loops linking underwriting, claims, service and prevention
Transaction trust	Accountability for autonomous action	Provide controls, auditability, liability transfer and assurance
Digital intelligence	Regulated risk-bearing capacity	Use the balance sheet as a strategic asset, not a commodity input

This framing reverses a common fear. AI could commoditize portions of insurance, particularly advice, administration and standard risk selection. Yet it may make the institution of insurance more important. Autonomous ecosystems need mechanisms for permissions, guarantees, liability allocation, recovery and compensation. Just as marine insurance supported trade under uncertainty and cyber insurance supported digital operations, new forms of agent, ecosystem and outcome assurance may support increasingly autonomous economic activity.

The insurance value chain becomes a learning system

The deepest implication is not the arrival of a new product category. It is the reconfiguration of the insurance value chain. Traditional insurance was optimized for episodic interaction and incomplete information. A customer was acquired, risk was assessed, a policy was issued, claims were handled and the contract was renewed. Each function could be managed as a specialized module. In an outcome ecosystem, information and decisions circulate continuously. Sensing informs prediction; prediction prompts intervention; intervention changes exposure; claims reveal whether the model and intervention worked; and that learning changes future design and underwriting.

This is where a new form of integration appears. It does not necessarily mean bringing every repair shop, clinic, vehicle platform or data provider onto the insurer’s balance sheet. It means integrating the feedback loops across distribution, underwriting, service, prevention and claims. A carrier that outsources all customer interaction, telemetry, intervention and claims learning may retain the policy and capital while losing the system that improves the outcome. In Sangeet’s formulation, the prize is not ownership of every asset but ownership of the learning connections between them.[1]

Value-chain element	Traditional model	Ecosystem and agentic model	Illustrative example
Demand and distribution	Customer or intermediary shops for a policy	Human or enterprise agent initiates an outcome journey and selects protection in context	Travel protection selected as part of a disruption-managed itinerary
Product design	Line-of-business policy with fixed terms	Modular protection, prevention, assistance and guarantees assembled around an outcome	Mobility assurance spanning accident, delay and recovery
Underwriting and pricing	Historical cohorts assessed at inception or renewal	Continuous, permissioned sensing with contextual assessment and explicit fairness controls	Driving feedback based on connected-vehicle measures
Policy administration	Contract-centric servicing and periodic changes	Event-driven entitlements, permissions and coverage adjustments	Coverage activated for a specific journey or asset state
Risk engineering	Separate advisory or loss-control service	Intervention embedded in the operating workflow	Maintenance or care prompt triggered before expected deterioration
Claims	Loss notification, validation, indemnification and recovery	Detection, prevention, automated evidence capture, rapid settlement and causal learning	Sensor-confirmed incident with pre-populated evidence and immediate assistance
Capital and portfolio	Pool largely managed by product and historical exposure	Portfolio steered using real-time accumulation, ecosystem dependencies and scenario simulation	Monitoring correlated mobility or infrastructure disruption
Learning	Insights remain inside functions and annual cycles	Closed feedback loops improve models, products, interventions and partner decisions	Claims outcomes update prevention rules and partner selection

Distribution therefore changes from persuading a customer to buy a product to earning eligibility inside an orchestrated decision. Insurers will need machine-readable products, transparent exclusions, reliable service levels and evidence that their protection improves the overall outcome. Brand remains important because people will not delegate consequential decisions to institutions they do not trust. But brand must be reinforced by technical discoverability, operational reliability and verifiable performance.

Underwriting changes from classifying a risk at a point in time to navigating how exposure evolves. That does not mean every premium should fluctuate continuously. Constant repricing can undermine solidarity, create customer anxiety and trigger regulatory concerns. The more important change is continuous risk understanding: identify emerging hazards, recommend interventions and reserve dynamic pricing for contexts where it is fair, explainable and permitted.

Case study: Cutting first-touch submission handling time by 58%

A commercial MGA was receiving roughly 250 submissions every week, most arriving as email packages that had to be reviewed manually before an underwriter could begin assessing the risk. The initial review took about 12 minutes per submission, which translated into roughly 50 staff hours every week spent simply opening files, locating required information, checking whether the package was complete and preparing it for the next step.

DICEUS first redesigned the intake process rather than placing AI on top of an inconsistent workflow. Required submission fields and routing rules were standardized, and one internal platform was established as the authoritative record. Only after that foundation was in place did we introduce AI-supported extraction and completeness checks to prepare incoming submissions for review. Underwriters continued to make the final risk decisions.

In practice, when a new submission arrived, the team no longer had to spend the same amount of time manually reconstructing the package before underwriting could begin. The workflow used the standardized data requirements and routing logic to prepare the submission in a consistent format, while the AI-supported layer handled extraction and initial completeness checking. The underwriter received a more structured submission and could focus on the actual risk assessment rather than the repetitive first-touch review.

The result was measurable. First-touch processing time fell from approximately 12 minutes to five minutes per submission. Across roughly 250 weekly submissions, total handling time dropped from about 50 hours to 20.8 hours per week – a 58 percent reduction. The improvement came not from AI alone, but from combining standardized data, a stable workflow, a single authoritative system and clearly defined human decision authority.

Claims changes from a downstream cost center into a strategic sensor. Claim files contain evidence about failure modes, repairability, customer behavior, partner performance and model error. AI can make intermediate work, rejected options and investigative trails machine-readable, turning what was previously operational exhaust into reusable learning capital.[1] This is one reason insurers should be cautious about outsourcing claims intelligence even when administration is outsourced. The by-product of handling the claim may become more valuable than the transaction itself.

Use case: Turning FNOL into structured claim evidence

The claims team faced a recurring problem at first notice of loss: before an adjuster could assess a case, they often had to spend time reconstructing it from customer descriptions, photos, supporting documents and policy information. The initial claim file could arrive incomplete or inconsistently structured, which meant the adjuster’s first task was frequently administrative – gathering and organizing evidence – rather than assessing the loss itself.

DICEUS redesigned the FNOL intake around a structured, AI-assisted workflow. At the first interaction, the flow captured the customer’s loss description and supporting photos, checked the relevant policy terms, organized the available evidence and prepared the initial claim record. It could also support an initial damage estimate and use predefined routing logic to separate straightforward cases from those requiring additional judgment. Complex cases were not decided automatically; they were escalated to an adjuster with the available evidence and policy context already assembled.

In practice, if a customer reported a simple loss and submitted the necessary description and photos, the case no longer had to begin with the same manual reconstruction process. The workflow organized the information at intake and prepared the claim for the next step. If the loss was unclear, the evidence was incomplete or the case required judgment, it was routed to an adjuster with a pre-filled file rather than a blank one.

The result was a more structured first-notice process and a clearer division of work between automation and human judgment. Routine intake and evidence organizations moved earlier in the workflow, allowing adjusters to spend more time on assessment and complex decisions. Straightforward claims could enter a faster handling path, while complex cases reached the adjuster with more context already available. The insurer also gained a structured claim record from the first interaction, creating a stronger basis for later claims analysis, underwriting feedback and prevention learning.

Finally, governance becomes part of the value proposition. Agentic systems require defined authority, human escalation, audit trails, cybersecurity controls and monitoring for drift. The WEF recommends progressive governance based on an agent’s role, autonomy, predictability and context.[8] Insurers can apply these controls internally, make them underwriting requirements for clients and potentially offer assurance services that assess whether autonomous systems operate within agreed boundaries.

Three strategic positions for insurers

The emerging architecture does not dictate a single winning model. It does, however, make the trade-offs clearer. An insurer can be a highly efficient capacity provider, a specialist risk-intelligence partner or an ecosystem trust orchestrator. Each can be viable, but drifting unintentionally into the first position is dangerous because agentic distribution may increase price transparency and reduce visible differentiation.

The capacity provider competes on capital efficiency, underwriting discipline and reliable APIs. This position can succeed at scale, but it accepts that another actor owns customer intent and the learning loop. The risk-intelligence partner contributes proprietary models, data and prevention capabilities to several ecosystems, earning value from better decisions without attempting to control the full journey. The trust orchestrator goes further: it helps define permissions, intervention rules, liability allocation, service guarantees and recovery across an ecosystem. This position is hardest to build, but it offers the strongest claim on future strategic relevance.

Insurers should choose deliberately where to lead, partner or supply capacity. A health insurer with care-management capabilities and longitudinal data may credibly orchestrate selected health outcomes. A commercial insurer with engineering expertise may become a risk-intelligence partner in industrial ecosystems. A reinsurer may supply capital, scenario models and governance frameworks across multiple orchestrators. The right position depends less on ambition than on access to context, authority to intervene, ability to measure outcomes and willingness to accept accountability.

A practical agenda: Build options before the market settles

Structural uncertainty makes fixed, five-year bets risky. Sangeet argues that capabilities should be treated as options that reveal new strategies, not only as execution assets for a predetermined plan.[1] Insurers should therefore build a portfolio of controlled experiments tied to different potential industry structures. One experiment might make products machine-readable for agentic distribution. Another might connect claims and prevention data around a narrow mobility or property outcome. A third might develop an assurance framework for enterprise agents. The purpose is not to run isolated pilots, but to test where the insurer can own a valuable feedback loop.

Three design disciplines matter. First, start with an outcome and a measurable failure mode, not with a technology. Second, instrument the full loop from recommendation to action to result, including cases where the recommendation is rejected. Third, establish data rights, consent, accountability and human override before scaling. McKinsey’s research similarly cautions that value depends on redesigning workflows and governance rather than adding agents to existing processes.[3][11]

Case study: Scaling captive operations through configurable workflows

The captive management operation had reached a point where growth was making its existing processes increasingly difficult to control. Policy issuance and renewals, endorsements, billing and invoicing, exposure and bordereaux data, reporting and document preparation still depended heavily on fragmented manual workflows. As the number of captives, policies, facilities and product variations increased, the same operating model created more duplicated work, greater risk of inconsistent data and more difficulty standardizing administration across the portfolio.

DICEUS began by mapping the existing operating processes rather than replacing them with a generic administration tool. The Captive Operating Platform was configured around the client’s actual model, including product and coverage structures, facility-level data, endorsement rules, billing logic, permissions, reporting requirements and document workflows. High-volume processes such as policy setup, renewals, endorsements, billing, invoicing, reporting, document templates, electronic signing and dashboard analytics were brought into one structured environment. The platform’s low-code configuration also allowed new captive-specific variations to be handled without hard-coding each one, while its modular architecture supported phased implementation without disrupting existing operations.

In practice, this changed how the team handled portfolio growth. Instead of creating another disconnected workflow when a new captive, policy structure or endorsement scenario appeared, the team could configure the required rules inside the same operating environment. For example, a new endorsement could be defined using the platform’s existing product structure, permissions, billing and document logic rather than being rebuilt as a separate manual process. The same shared data could then flow through policy administration, billing, reporting and owner-facing workflows instead of being recreated across spreadsheets and email-based approvals.

The result was a single operating environment that now supports 55 captives, more than 3,000 policies and over 100 custom endorsements. The implementation replaced fragmented manual administration with structured and automated workflows, created a single source of operational data, improved process consistency and visibility, and reduced the risk of data errors. Just as importantly, growth no longer required rebuilding the operating model for every new variation: the same configurable architecture could absorb additional captives, policies and endorsement scenarios as the portfolio expanded.

Leaders should also ask what must remain integrated. If claims evidence improves underwriting, if prevention interventions alter product economics, or if partner performance determines customer outcomes, those feedback pathways require common measurement and governance even when execution is distributed. Conversely, activities that do not contribute to distinctive learning may remain modular or be sourced externally. The strategic boundary of the insurer should be drawn around the loops it must learn from, not simply around the activities it has historically owned.

The strategic question – Trust orchestration

For the past decade, insurers have asked how to distribute products through super apps, digital platforms and embedded propositions. That was the right question when the challenge was how to reach customers at the point of transaction. The next phase asks something more fundamental. As intelligent systems increasingly interpret intent, compare alternatives, coordinate providers and act on behalf of customers, insurers need to consider not only how their products will be distributed, but where they want to sit in the decision and risk architecture around the customer. In other words: when AI increasingly coordinates the journey, what role does the insurer need to play to remain relevant to the outcome?

The answer starts with a broader view of what insurance can provide. A policy that pays after a loss remains fundamental; risk transfer is not becoming obsolete. But the capabilities around that policy can become much more valuable. In more connected and increasingly autonomous ecosystems, insurers can help identify emerging risk, support prevention, define boundaries for autonomous action, absorb residual uncertainty and help customers recover when the system fails. Claims, underwriting, assistance, prevention and risk engineering can become connected parts of the same learning loop rather than isolated moments in the insurance lifecycle. In that sense, the policy does not disappear. It becomes part of a wider promise of resilience and accountability.

This does not mean that every insurer needs to become an ecosystem orchestrator. Nor does it suggest that orchestration is automatically the most attractive position. Some insurers will create value as highly efficient capacity providers; others through specialist risk intelligence, prevention or assurance. A smaller group may have the customer access, data, authority and capabilities to coordinate a broader outcome. What matters is choosing that position deliberately – and understanding which data, decision rights, customer relationships and learning loops are too important to give away.

The strategic issue is therefore not whether every insurer should try to control the ecosystem. It is deciding which parts of the emerging risk architecture it needs to own, where it should partner and where it is comfortable becoming a supplier to someone else’s journey. That choice has consequences. An insurer that gives up customer context, claims intelligence or the ability to influence risk may gain efficiency while gradually losing the feedback loops that make its proposition distinctive. Conversely, trying to orchestrate an ecosystem without the necessary customer permission, operational authority or ability to improve outcomes can add complexity without creating value.

As AI takes on more coordination and decision-making, these choices become more consequential. Insurance has a credible role because the industry already combines capabilities that autonomous ecosystems will continue to need: risk understanding, prevention, capital, contractual certainty, recovery and accountability when things go wrong. The opportunity is to connect those capabilities in ways that are useful before, during and after the moment of loss.

For business leaders, the question is therefore no longer simply, “What is our AI strategy?” A more useful question is: “Where do we want to sit when AI increasingly orchestrates the customer journey – and what must we understand, control and learn from to deserve that position?” The answer will differ by insurer and by market. But making that choice deliberately may become one of the defining strategic decisions of the next phase of insurance.

About the authors

Illia Pinchuk is the CEO of DICEUS, a technology company specializing in software solutions for insurers, brokers, MGAs, and alternative risk organizations. He leads the company’s strategy and product development, focusing on insurance technology modernization, AI-driven automation, and digital transformation. His work centers on helping insurance organizations simplify complex operations, improve efficiency, and adopt scalable technology solutions.

LinkedIn: Illia Pinchuk

Dr. Evangelos Avramakis is an expert in strategic foresight, innovation, artificial intelligence, and the future of insurance. With over 20 years of experience in strategy and business transformation, he previously served as Lead Foresight, Intelligence & Development at Swiss Re. Today, he advises organizations on navigating disruption, applying AI, and developing future-ready business models.

LinkedIn: Evangelos Avramakis

Sources

Sangeet Paul Choudary, “Strategy in the Age of AI: Eight Points Beyond the Obvious,” Platforms & AI, 2026.
Grab Holdings, “Grab Files Annual Report on Form 20-F for Fiscal Year 2025,” March 7, 2026.
McKinsey & Company, “The Agentic Organization: Contours of the Next Paradigm for the AI Era,” September 26, 2025.
Marc Hasselwander, “Digital Platforms’ Growth Strategies and the Rise of Super Apps,” Heliyon 10(5), 2024.
Vitality and RAND Europe, “Financial Incentives and Physical Activity,” summarized by Vitality, November 28, 2018.
Tesla, “Safety Score,” accessed August 2026.
Neha Agarwal and David Basson, “The Power of Two: Digital Twins in Insurance,” The Actuary, April 4, 2024.
World Economic Forum and Capgemini, “AI Agents in Action: Foundations for Evaluation and Governance,” November 2025.
Deloitte Insights, “From Manufacturing to Medicine: How Digital Twins Can Unlock New Industry Advantages,” June 5, 2025.
World Economic Forum, “How to Design for Trust in the Age of AI Agents,” February 18, 2026.
McKinsey & Company, “Five Fifty: The Agentic AI Opportunity,” 2025.
Deloitte, “2026 Global Insurance Outlook,” October 9, 2025.