What’s Stopping Organisations From Trusting Their Own AI?

A senior leader at a global pharmaceutical company said it plainly. They weren’t lamenting a lack of investment. They weren’t waiting on a new model or platform. The data exists. The tools exist. What’s missing — across industries, geographies, and organisational sizes — is trust.

That single observation, drawn from a new research report by Tech Eco Hub and Coveo, captures the central challenge facing enterprise AI in 2026. Organisations are not failing to adopt AI. They’re failing to trust it. And the reasons why are more structural, more human, and more tractable than most technology strategies acknowledge.

When AI Exposes What Was Already Broken

A common misconception about AI search and discoverability is that it solves information problems. In practice, the research found the opposite: AI amplifies pre-existing weaknesses in how organisations structure, govern, and connect their knowledge.

Fragmented systems, inconsistent data definitions, and unclear ownership — issues that were once tolerated — now directly limit the effectiveness of AI initiatives. When an AI system encounters conflicting definitions of the same data across different business units, it doesn’t resolve the ambiguity. It scales it.

Relevant reading: Clean Data, Smarter AI: The Critical Role of Knowledge Management in Maximizing Generative AI

As Stephen Page, Domain Product Owner at Experian, explained: the core problem isn’t a shortage of data — it’s the absence of shared understanding about what that data represents. When different parts of a business interpret the same information differently, AI will faithfully reproduce and amplify that confusion at speed.

Discoverability Is Now a Governance Problem

The shift from keyword-based search to intent-based, agentic AI discovery changes what discoverability actually means. AI systems no longer retrieve and rank content — they infer, synthesise, and generate responses based on the connected knowledge available to them.

This has a critical implication: if the underlying knowledge is poorly structured, siloed, or semantically inconsistent, AI systems become more likely to hallucinate — producing confident-sounding but inaccurate responses. In regulated industries, this isn’t just an inconvenience. As the same pharmaceutical leader noted, in an environment dealing with intellectual property and clinical information, the wrong answer is simply not an option.

The research makes clear that discoverability can no longer be treated as a UX feature or a search relevance problem. It is a data governance challenge. Organisations must be able to trace where AI-generated answers come from, who owns the underlying data, and how its meaning has been defined — not just for compliance, but for the basic purpose of trusting the output.

Relevant reading: 6 Data Cleaning Challenges Blocking Enterprise AI (& Solutions)

The Organisational Gap Is Wider Than the Technology Gap

Across interviews with senior leaders at Vodafone, Experian, P&O Cruises, a global energy company, a multinational publisher, and a global pharmaceutical business, one theme surfaced consistently: the hardest part of AI readiness is not technical. It’s organisational.

Business units operating with separate data ecosystems, ownership disputes that prevent data sharing, governance structures that centralise control without enabling local teams — these are the real blockers. Technology can be purchased. The organisational alignment required to make it work cannot.

Relevant reading: How The Silo Problem Is Killing Your Customer Service Experience

The AI Product Owner at a Multinational Energy Company captured it plainly: bringing data together across business units is not technically difficult. The difficulty is organisational — persuading people to share, standardise, and take collective ownership of something they’ve always treated as their own.

The Cost of Poor Data Hygiene Is No Longer Abstract

The research identifies four concrete consequences that follow when data foundations aren’t strong enough to support AI — and at least one of them directly undermines the very reason most organisations deployed AI in the first place.

These aren’t theoretical risks. They’re the daily operational reality for organisations that have moved fast on AI deployment without addressing the data foundations beneath it.

Building Trust Incrementally — Not Overnight

The research is clear that trust in AI systems cannot be rushed. The leaders who are making progress share a common approach: start small, instrument carefully, learn continuously, and expand only when confidence is earned.

Jitendra Mulchandani, Senior Product Manager at Vodafone, summarised the mindset simply: try, watch, learn, then implement. It’s not a cautious philosophy — it’s the only one that builds the kind of trust that survives contact with a real operational environment.

Em North, Digital Marketing Manager at P&O Cruises, adds an important qualification: the ambition to move is there, but so is the caution — especially when technology is outpacing the governance structures designed to keep it in check. The challenge for most organisations isn’t a lack of will to act, but ensuring the frameworks around AI can keep pace with the speed at which it’s being deployed.

What This Means for Leaders Right Now

The report outlines five priorities for the next 12 to 24 months — including one organisational change that several companies said was the hardest to implement.

In the end, the research reaches a striking conclusion: AI does not replace organisational intelligence. It reveals whether it truly exists.

The full report draws on candid conversations with senior leaders at Vodafone, Experian, P&O Cruises, and several other large organisations across energy, publishing, and pharma. It explores how the definition of search is changing, what data foundations AI actually requires, and the practical steps organisations should prioritise over the next 12–24 months.