AI is exposing governance problems that have been there for years. For many insurers, that’s the real reason progress is slowing today.
Across the insurance industry, organisations are investing heavily in AI, from claims automation and fraud detection to underwriting optimisation and customer service. The opportunity is clear.
But many initiatives stall long before they deliver value. And it’s rarely the technology holding things back. It’s the data.
AI doesn’t create data problems. It brings them into the open, quickly and unavoidably.
The same issues that have long caused reporting discrepancies, regulatory friction and operational inefficiencies don’t disappear when AI is introduced. In insurance, the consequences are very real: inconsistent policy or claims data affects reserving accuracy; poor bordereaux quality leads to reconciliation issues with capacity providers; undocumented manual processes distort underwriting MI.
Layer AI on top, and those issues don’t improve, they scale.
For data leaders, this typically shows up in familiar ways:
- AI use cases taking longer than expected to reach production
- Teams spending more time validating outputs than acting on them
- Confidence in data being challenged at the point it matters most, often in front of senior stakeholders or regulators
So the question is simple:
Are your data foundations strong enough to support AI at scale, or are they quietly holding it back?
Are your data foundations strong enough to support AI at scale, or are they quietly holding it back?
This health check is designed to help insurance leaders sense-check whether governance foundations are genuinely fit for AI and where to focus first to unlock value faster.
1. Do people actually trust the data?
Most insurers have governance frameworks, policies and documentation. Far fewer have genuine trust in their data.
A simple sense-check: when performance figures are presented in a meeting, does the conversation move quickly to action, or drift into debating which number is right?
In insurance, this matters. If underwriters question the MI, if claims handlers rely on their own spreadsheets, or if actuaries adjust figures before submission because they don’t trust the source, governance isn’t working in practice, regardless of what the framework says.
Health check
☐ Business users trust the data used for reporting and decision-making without needing to verify it independently
☐ Data is used with confidence in automation and AI use cases, not worked around
☐ Teams are not routinely asking which number is correct or maintaining parallel versions
☐ Data is used with confidence in automation and AI use cases, not worked around
☐ Teams are not routinely asking which number is correct or maintaining parallel versions
How many of these 3 can you answer yes to with confidence? _____ / 3
Reality check
If stakeholders are still validating outputs instead of acting on them, AI will struggle to create value.
2. Is data ownership genuinely accountable?
Many organisations can point to a list of named data owners. The real question is whether those individuals understand what they own, why it matters, and what happens if things go wrong.
In practice, ownership is often assigned based on hierarchy or system responsibility rather than true accountability. In the London Market and Lloyd’s environment, this becomes even more complex: data about a single risk may pass through multiple parties before reaching the carrier.
Where ownership breaks down, quality issues follow, and they’re hard and costly to fix.
Health check
☐ Data owners understand specifically what they own and why it matters to the business
☐ Ownership is embedded in role responsibilities, not just recorded in a governance register
☐ Non-compliance with ownership obligations has real consequences — reputational, operational or otherwise
☐ Ownership assignments are reviewed when organisational or system changes occur
☐ Ownership is embedded in role responsibilities, not just recorded in a governance register
☐ Non-compliance with ownership obligations has real consequences — reputational, operational or otherwise
☐ Ownership assignments are reviewed when organisational or system changes occur
How many of these 4 can you answer yes to with confidence? _____ / 4
Reality check
If ownership disappears when issues arise, it was never really there. Documentation doesn’t change behaviour, accountability does.
3. Does governance reflect how the business actually operates?
Governance programmes often describe how things should work. The reality is different.
Over time, manual interventions, spreadsheet workarounds and local fixes creep in to keep things moving, often without being documented or understood beyond the team that created them.
In insurance, this shows up clearly in claims handling, bordereau reconciliation and regulatory reporting. When key individuals leave, or when AI is introduced, those gaps surface quickly.
Health check
☐ Governance reflects how work actually happens, not just how it was designed
☐ Manual and spreadsheet-based processes are visible, documented and actively managed
☐ Key process knowledge is held in systems and documentation, not just in people’s heads
☐ Shadow processes are being reduced, not tolerated
☐ Manual and spreadsheet-based processes are visible, documented and actively managed
☐ Key process knowledge is held in systems and documentation, not just in people’s heads
☐ Shadow processes are being reduced, not tolerated
How many of these 4 can you answer yes to with confidence? _____ / 4
Reality check
If knowledge sits with individuals rather than systems, AI will expose that risk immediately.
4. Are data quality efforts focused on business value?
A common trap is trying to fix everything. That typically leads to long backlogs, limited impact and governance fatigue.
The organisations making real progress are deliberate about where they focus, prioritising based on customer impact, operational efficiency, regulatory risk and commercial value.
In insurance, the highest-value areas are well understood: regulatory reporting accuracy, claims leakage, bordereaux quality and underwriting MI. These are where data issues translate directly into measurable business outcomes.
Health check
☐ Data quality issues are prioritised based on business impact and regulatory risk, not just volume
☐ Teams can clearly articulate which issues have the greatest operational or commercial impact
☐ Data quality activity is tied to measurable outcomes, not just ticket volumes
☐ There is a clear process for escalating issues that affect regulatory submissions
☐ Teams can clearly articulate which issues have the greatest operational or commercial impact
☐ Data quality activity is tied to measurable outcomes, not just ticket volumes
☐ There is a clear process for escalating issues that affect regulatory submissions
How many of these 4 can you answer yes to with confidence? _____ / 4
Reality check
If everything is treated as critical, nothing is. Without prioritisation, value is hard to demonstrate and even harder to scale.
5. Are controls proactive rather than reactive?
Many controls have been introduced in response to specific events — a regulatory finding or an audit observation. Over time, this creates layers of controls that add cost and complexity without improving risk management.
More effective governance takes a deliberate approach: designing controls around data flows, risk exposure and outcomes from the outset.
Health check
☐ Controls are designed around data risk and movement, not added retrospectively
☐ Preventative controls are balanced with detective controls
☐ Duplication, gaps and redundant controls are actively assessed and reduced
☐ Controls across third-party and delegated authority data are as robust as internal ones
☐ Preventative controls are balanced with detective controls
☐ Duplication, gaps and redundant controls are actively assessed and reduced
☐ Controls across third-party and delegated authority data are as robust as internal ones
How many of these 4 can you answer yes to with confidence? _____ / 4
Reality check
If controls exist simply because they’ve always been there, they may be adding cost rather than reducing risk.
6. Is governance embedded in AI and transformation programmes?
Governance is most effective when it’s built in from the start, not added at the end.
The insurers scaling AI successfully embed governance into programme design, architecture and delivery. Data lineage is understood early. Critical data elements are defined upfront. Ownership is clear before change begins.
The opposite is also true. When transformation runs without governance, progress can quickly undo itself.
Health check
☐ Governance is built into AI and transformation programmes from the outset
☐ Change teams actively use governance artefacts such as data dictionaries and ownership registers
☐ AI initiatives are built on data that is trusted and well understood
☐ Governance standards are maintained as systems and processes change
☐ Change teams actively use governance artefacts such as data dictionaries and ownership registers
☐ AI initiatives are built on data that is trusted and well understood
☐ Governance standards are maintained as systems and processes change
How many of these 4 can you answer yes to with confidence? _____ / 4
Reality check
If governance only appears at implementation, it’s already too late.
Your AI readiness score
Add up the number of boxes you could answer yes to with confidence — not in principle.
20–24 | Strong foundations
You’re well positioned to scale AI with confidence. The focus now is on sustaining and evidencing what you’ve built.
You’re well positioned to scale AI with confidence. The focus now is on sustaining and evidencing what you’ve built.
13–19 | Making progress
There are gaps that may limit the value of AI. Prioritise where scores are weakest to unlock impact faster.
There are gaps that may limit the value of AI. Prioritise where scores are weakest to unlock impact faster.
Below 13 | Action needed
AI initiatives are likely to face significant friction until foundations are strengthened. Start with ownership and trust.
AI initiatives are likely to face significant friction until foundations are strengthened. Start with ownership and trust.
Final thought
The insurers seeing the most value from AI aren’t always those with the most advanced technology. They’re the ones who trust their data, understand how their processes really work, have clear accountability, and embed governance into how they operate day to day.
That’s what turns AI from experimentation into real, scalable business impact.
Before asking whether your organisation is ready for AI, it’s worth asking a simpler question:
Is your data ready first? Book our Power Hour to get your AI on track, here, or watch our webinar where we unlock the reasons why governance still isn’t working.