Dom Selvon is Value Partner - CTO, at Valliance. His full argument was first published as Why trust, not technology, is holding enterprise AI back in Enterprise Times on 14 January 2026.
The biggest barrier to enterprise AI returns is trust, and it sits inside the organisation rather than the technology. UK enterprises are spending £39.2 billion on AI, according to our research, and the returns are lagging the spend. The usual suspects get blamed, whether that is data quality, integration complexity or a skills gap. Our CTO Dom Selvon, writing in Enterprise Times, put the real bottleneck elsewhere. Until people trust what AI is doing, no amount of technical capability converts into value.
The trust deficit runs through every level
The deficit looks different depending on where you sit. A CEO trusts AI's potential to transform operations, and doubts the organisation's readiness to deliver it. A CIO trusts the architecture, and doubts the data that powers it. On the ground, people resisting AI are usually acting on legitimate personal concerns, from job insecurity to the fear of being found wanting next to an AI output.
Each of these is rational. Together they are more damaging to returns than any technical shortfall, because a system nobody acts on returns nothing. As Dom wrote in the piece,
"AI can outpace any analyst, predict market shifts and even automate intricate workflows. Yet, when executives hesitate to act on its recommendations, or teams quietly bypass automated systems, that potential evaporates."
Trust can be engineered
The mistake is treating trust as something that grows organically once the technology proves itself. It doesn't. Organisations that wait for it lose ground to those building their people's readiness deliberately.
The engineering starts with visibility. Make the AI systems transparent, and set out from day one the outcomes they will be measured against, so everyone works from the same picture. Most programmes skip this step. Our research found just 45% of AI projects have success metrics laid out from the outset.
Training carries the other half. Skills alone don't produce usage; people can be fluent in a tool and still choose not to rely on it. Training has to build confidence in the tools alongside competence with them, otherwise the licence spend leaks straight out of the programme.
Value-first use cases build trust fastest
What you point AI at shapes how far people trust it. Simple, easy-to-deploy applications feel safe, and they also feel small. If people cannot see the technology amplifying what they do, they file it under tooling rather than transformation.
Choosing use cases for disproportionate impact works harder. Start where data assets are underused, even where data quality is currently poor, and involve the whole C-suite in the choice, because the technology function alone cannot see every division's biggest needs. Delivered well, one high-value use case does more for belief in AI than a dozen safe pilots. The CEO gets investment concentrated where the board will feel it. The CIO gets technology decisions anchored to named business goals. And the people using it get proof, in their own work, of what the technology can carry.
What trust makes possible
Too much AI today is deployed as a veneer, picked up ad hoc and never fully embedded in operations. That caps the short-term results and feeds scepticism about whether the technology deserves its billing.
The future enterprise has AI woven through its processes, making decisions faster, more accurate or easier to take, with a shared ontology connecting capability to operations and unified governance keeping it in check. None of that holds up on a workforce that quietly routes around the systems.
People accept AI as intrinsic to their work once they are convinced of what it can do today and assured they are under no threat for using it. Building that assurance deserves the same seriousness as any technical decision. Enterprises choosing a partner should weigh it the same way, and favour those that put trust and measurable outputs over flash.
