Who Controls Enterprise AI? CDO Vision London 2026

As AI adoption accelerates, so do the questions around control. Organizations are not just building AI systems, they’re figuring out how to govern them, trust them, and ensure they don’t create more risk than value.

At a recent CDO Vision Global AI Series event held in London, The panel featured Roxanne Howdle-Rowe (British Business Bank), Effie Kilmer (Microsoft), and Joel Lange (Dow Jones). CDO Vision London brings together the world's most forward-thinking data and technology leaders for candid, high-impact conversations about the challenges shaping modern enterprises. What emerged was a refreshingly grounded conversation about culture, trust, licensing.

Thousands of potential use cases, competing stakeholder priorities, and boards demanding visible progress, it's a recipe for paralysis. But according to leaders on the ground, the starting point is simpler than it seems.

"It always comes down to two buckets," noted Effie Kilmer. Organisations are either looking to reduce costs or grow revenue, and every AI initiative ultimately maps to one of those two outcomes. The challenge is deciding which use cases to pursue first.

That's where data readiness becomes the critical differentiator. Many organisations build ambitious AI roadmaps, only to discover that their underlying data infrastructure can't keep pace with the targets they set. Fragmented data sources, inconsistent quality, and poor lineage documentation all act as invisible brakes on AI momentum.

No implementation plan survives first contact with a workforce that's afraid. And right now, across almost every industry, that fear is tangible.

Roxanne, Managing Director of Data and Analytics at British Business Bank, described the moment many leaders will recognise, the introduction of a new AI tool met not with curiosity, but with a quiet, unmistakable look of anxiety. The concern isn't irrational. Layoffs across big tech, consulting, and financial services are making headlines, and employees connecting the dots between automation and redundancy are doing exactly what you'd expect rational people to do.

But the panel's leaders were clear: ignoring the fear doesn't make it go away. Acknowledging it openly and then actively countering it with transparency and time is what moves the needle. 

Roxan described building in deliberate test-and-learn periods, running AI tools in parallel with existing processes rather than as immediate replacements, giving colleagues space to engage without feeling immediately threatened.

Kilmer drew a longer historical arc, comparing today's anxiety to the disruption caused by the automobile, and closer to home, the arrival of Microsoft Excel. Both technologies displaced certain tasks while simultaneously creating entirely new categories of work and expertise. The mindset shift required, she argued, must come from the top down, supported by internal champions who model curiosity rather than resistance.

The emerging framing gaining traction across organisations positions AI not as a replacement but as a force multiplier, handling volume and repetition so people can focus on judgment, creativity, and relationships. Skilling programmes, peer champions, and honest communication about timelines are proving far more effective than mandates alone in bringing workforces along on the journey.

If the first wave of enterprise AI was about excitement, the second is about accountability. And for many organisations, that accountability conversation is arriving faster than their governance frameworks can handle.

The challenge is multidimensional. On one side, internal governance, ensuring employees use approved tools, do not feed sensitive documents into external large language models, and that every AI use case has cleared the appropriate organisational checkpoints. On the other hand, external governance, particularly for data publishers and aggregators, where the question of who owns content and who has the right to train models on it, has become deeply contentious.

Joel Lang framed this tension. As a major content publisher, Dow Jones has pursued what its CEO describes as a dual strategy of "wooing and suing", striking licensing deals with players like OpenAI while simultaneously pursuing legal action against those deemed to be misusing proprietary content. The underlying principle is straightforward: credible journalism should be paid for, and any AI system surfacing that content should have a legitimate, traceable licensing arrangement underpinning it.

For organisations consuming AI outputs, the same logic applies in reverse. Knowing where your data comes from, whether it's licensed, and whether the model drawing on it has handled that content responsibly is no longer a nice-to-have; it's a baseline requirement for trust.

Legal teams, once peripheral to technology decisions, are now central. Lang noted how Dow Jones established a dedicated AI committee spanning editorial, legal, and data leadership to review external use cases, a process that was initially slow and cumbersome, but has become significantly more efficient as both internal teams and customers have built familiarity with the key questions around data residency, model selection, and security.

Perhaps most essentially, the panellists agreed on one point: poor data quality not only slows AI down, but also actively undermines it. As Lang put it, deploying AI on top of badly structured, poorly governed data produces outputs that are inaccurate, unreliable, and potentially damaging. A strong data strategy doesn't just support AI. It determines how good the AI can actually be.