Perspectives

Preparing insurance for agentic AI: the urgent case for data modernisation

Agentic AI is moving from experimentation to operational reality. The question for insurers is whether their data is ready for it.

Even the cleverest pilots will struggle if your data estate isn’t prepared for autonomous action. We're seeing similar patterns in insurance to those emerging in asset management, but with an even sharper data challenge.

At the same time, insurers face growing pressure from rising customer expectations, increasing competition, regulatory scrutiny, and the need to reduce costs while maintaining control.

Together, three forces are pushing data modernisation to the top of the agenda.

1. AI is moving from insight to action

Traditional AI, such as predictive machine learning models and rules-based automation, has helped insurers to analyse risk, detect anomalies, triage claims, and support human decision-making. Agentic AI goes further to:

Orchestrate entire claims journeys
Support dynamic underwriting
Coordinates decisions across systems
Execute complex steps across the insurance value chain, such as routing claims, checking policy conditions, triggering referrals, requesting missing evidence, and initiating customer communications.

Salesforce describes agentic AI in insurance as a shift from AI that “suggests” to AI that “acts”, while DXC describes a new agentic age of intelligent autonomous systems that can perceive, reason, act and learn. That shift introduces a hard requirement: AI must operate on trusted, real-time, connected data, not fragmented legacy silos. Microsoft argues that insurers can augment and accelerate core platforms through targeted, extensible AI capabilities, but that doing so depends on connected workflows, integration with core systems, and a unified data platform.

2. Legacy data estates are the bottleneck

Most insurers still rely on a combination of:

Siloed policy administration systems
Claims platforms with inconsistent data models
Document-heavy workflows
Manual workarounds and spreadsheet-based processes
Inconsistent, duplicated, or incomplete customer and risk data

This creates a structural problem. AI cannot access the full context it needs. Automation breaks across system boundaries. Scaling beyond pilots becomes difficult because the data is not reliable, accessible, or governed enough to support operational use. The issue is not that insurers lack data. It is that too much of its data is trapped in systems, documents, and processes that were designed for human interpretation rather than machine reasoning.

3. Data complexity is increasing

Insurance data is inherently complex:

Structured data, such as policies, claims, pricing, billing, and exposure records
Semi-structured data, such as broker submissions, proposal forms, bordereaux, and schedules
Unstructured data, such as documents, images, emails, call transcripts, loss adjuster reports, and medical or engineering records

At the same time, insurers are expected to:

Personalise products and service journeys
Make faster and more consistent decisions
Integrate third-party and ecosystem data sources
Support explainability, auditability, and regulatory oversight
Move from retrospective reporting to real-time operational intelligence

This dynamic requires a shift from fragmented, static data to dynamic, orchestrated, and governed data ecosystems. Business Insider's analysis of agentic AI data foundations frames this as three shifts: static to dynamic, siloed to orchestrated, and reactive to proactive.

Why data modernisation - not system replacement - is the priority

Many transformation programmes still focus first on core system replacement. But in the context of AI, that is often the wrong starting point. The real constraint is whether the organisation can make its data usable, governed, and accessible enough for AI to support and eventually execute real workflows. Leading insurers are therefore focusing on:

Data unification across silos
Real-time data access, not batch-only reporting
Clear ownership of critical data domains such as customer, policy, claim, risk, and exposure
Strong governance, quality, lineage, and auditability
API-first and composable architectures that allow systems to evolve incrementally
Data products and reusable services that support multiple AI use cases

This approach creates a foundation where AI models can reason effectively, agents can act safely, and workflows can be orchestrated end-to-end. BCG makes a similar point in the context of insurance IT modernisation: core modernisation is complex because of massive data sets, system dependencies, and jurisdictional variations, but agentic AI and stronger design discipline can help cut through that complexity.

4 ways to prepare for agentic AI

Agentic AI introduces a step change in requirements. It is not simply another analytics layer, chatbot, or productivity tool. It changes the operating assumption from AI as an assistant to AI as an active participant in workflows.

Traditional AI	Agentic AI
Supports decisions	Executes decisions
Uses curated datasets	Requires enterprise-wide data
Works in isolation	Operates across systems
Low autonomy risk	High governance requirement

To support this, insurers must evolve their data strategy in four practical ways:

1. From siloed to connected

Agents need access to the full customer, policy, claim, risk, and interaction context. A claims agent, for example, cannot act safely if it can see the FNOL record but not coverage, exclusions, fraud indicators, payment history, prior claims, or current customer vulnerability flags.

2. From batch to real-time

Decisions must happen in-flight. Claims triage, underwriting referral, fraud checks, renewal outreach, and customer servicing all depend on current data. Data that is accurate only after an overnight batch window will limit what agents can safely do. An underwriting agent, for example, cannot make an in-flight decision if broker submissions, claims history, exposure data, sanctions screening, pricing appetite, and capacity constraints are only refreshed overnight rather than available at the point of decision.

3. From passive to governed

Data must be continuously validated, explainable, and auditable. As agents gain autonomy, governance must move from periodic review to embedded control: permissioning, monitoring, lineage, escalation thresholds, and clear ownership. An AI agent making a claims or underwriting recommendation, for example, cannot be trusted if the insurer cannot trace which data it used, whether that data was approved, when it was last validated, and whether the decision complied with underwriting, claims, conduct, and regulatory controls.

4. From human-mediated to machine-consumable

Data structures must support autonomous reasoning, in addition to human reporting. That means improving metadata, standardising key business entities, extracting knowledge from documents, and making rules, policies, and operational context available in forms that agents can interpret reliably. An agent cannot reliably reason over policy wording, exclusions, endorsements, broker notes, claims correspondence, and loss adjuster reports if that information remains locked in PDFs, email threads, spreadsheets, or free-text notes that require a human to interpret them.

The cost of waiting

Delaying data modernisation has compounding effects:

AI programmes remain stuck in pilots, and the gap between what the business wants AI to do and what the data estate can support keeps widening
Operating costs rise as manual workarounds become harder to unwind and legacy complexity becomes more expensive to address
Customer experience falls behind digital-native competitors while regulatory and audit exposure quietly grows

Meanwhile, insurers that invest early are better positioned to scale AI use cases, reduce transformation risk, and unlock faster returns. McKinsey's 2025 insurance AI report argues that insurers need to move beyond pilots and rewire business domains, operating models, data, and technology to capture meaningful value from AI.

Where to start?

The good news is that you don’t need to invest in a broad, enterprise-wide data modernisation programme. Rather, it is a sharper set of diagnostic questions that help identify where data is already constraining performance and where AI could create measurable value. Insurers should ask:

Which journey in the business is most constrained by data quality today? For example, claims triage, underwriting, renewals, bordereaux processing, complaints, fraud, or customer servicing.
Where would trusted, real-time data most improve customer, risk, or operational outcomes? This helps prioritise journeys where better data foundations could reduce leakage, speed up decisions, improve service, or strengthen control.
Which decisions or workflows could benefit from agentic AI, but are currently held back by fragmented systems, unstructured documents, or unclear data ownership? This shifts the conversation from "where can we use AI?" to "where is the business ready—or not yet ready—for AI to act safely?"
What governance, controls, and integration would need to be in place before an AI agent could operate safely in that journey? This is where insurers move beyond experimentation and start thinking about trust, auditability, accountability, and production readiness.

These questions help insurers avoid modernisation for its own sake. The aim is to focus investment where stronger data foundations can identify real business value and create the conditions for AI and agentic AI to scale responsibly.

Insurance has reached an inflection point

The trifecta is clear: AI is moving from insight to action, legacy data estates are constraining progress, and data complexity is increasing. Together, these forces make data modernisation a strategic priority.

AI—particularly agentic AI—will change how insurers operate. But its success will not be determined by model sophistication alone. It will depend on whether insurers have the data foundations needed for AI to reason, act, and be governed safely.

The cost of waiting is that friction becomes a constraint: pilots remain difficult to scale, manual workarounds become harder to unwind, and fragmented data limits speed, service, and control. The winners are looking beyond fast experimentation to creating fertile conditions for AI to scale responsibly.

That means having the right data, in the right shape, at the right time. For most insurers, that means the time to modernise data is now.