Shaping data mindsets for an AI-ready future

Why organisational honesty, not technology, will determine AI success

The conversation about data and AI has shifted from excitement to reckoning. The question is no longer whether organisations should pursue generative or agentic AI, but whether they are being sufficiently honest about the state of their data foundations to do so responsibly. What emerges from executive-level cross-sector dialogue is a revealing truth: the biggest gap is not technological capability, but organisational candour.

What HotTopics and Confluent have learned over the course of a year hosting C-suite debates on this crucial topic is this: when executives are asked to rate their data maturity against their AI ambitions, the answers are rarely binary. Instead, they fragment. And although this introduces severe complexity in a system already difficult to manage, it makes for fascinating and timely discussions.

Technology and data leaders enjoyed this kind of discussion at a recent Food for Thought lunch, hosted at London’s St Paul’s Cathedral. Their combined insights can be read below. 

Tensions at the heart of business architecture

Back to fragmentation. The roundtable heard live transactional data may be robust, while static or customer data lags behind. External-facing products may be far more advanced than internal decision-making; pockets of excellence coexist with structural weakness. This unevenness matters, because AI does not operate selectively. It consumes whatever it is fed, amplifying both strengths and flaws at scale.

One of the clearest patterns to emerge is the growing discomfort with overconfident narratives. 

Many organisations have invested heavily in modern data architectures, data mesh concepts and AI tooling, yet privately acknowledge that the organisational model has not kept pace. Building “data products” without reorganising accountability around business domains creates a fragile equilibrium that is technically elegant but managerially unresolved. Several leaders admitted, with refreshing candour, that they are laying infrastructure that their own operating model is not yet capable of sustaining.

This tension surfaces most sharply around data ownership. 

Despite years of discussion, many enterprises still conflate ownership with custody, and responsibility with access. IT teams manage platforms, business teams generate and consume data, and everyone assumes someone else is accountable for quality. The moment senior leaders are formally named as data owners, resistance appears. Ownership, once accepted, forces uncomfortable follow-on questions: who uses my data, how good is it, and what decisions depend on it? For many executives (outside of data-focused ones), this is the first time data ceases to be abstract and becomes managerial.

• “What do we allow?” 
• “Where do we draw boundaries?”,“
• "What are the reputational and regulatory consequences if this fails?”.

Only once those guardrails were clear did they confront the harder realisation: their most valuable use cases were constrained not by algorithms, but by fragmented, inconsistent or poorly owned data.

This is where data foundations reassert themselves as a strategic capability.

Leaders increasingly reject the idea that data quality must be perfect everywhere. Instead, they practise “conscious prioritisation”. Data that drives regulatory reporting, customer decisions or material financial outcomes is treated differently from data that is informational or peripheral. Investment follows impact. Quality is improved where it matters most, while imperfection is acknowledged and managed elsewhere. This is not complacency; it is realism.

Several participants described this as continuous data triage. 

As business priorities shift, so too does the data that deserves attention. This requires not just tooling, but literacy. Business leaders must understand what data can and cannot support, what confidence levels are acceptable, and where automation introduces risk rather than efficiency. In this context, AI becomes less a destination and more a stress test. It reveals weak semantics, unclear definitions, and brittle pipelines that humans have quietly worked around for years.

AI ambition vs. enablement

Another recurring theme was the imbalance between ambition and enablement. 

Many organisations are launching AI training programmes, encouraging experimentation and signalling bold intent. Yet internally, data teams remain underpowered, under-permissioned, or trapped between competing owners. Analysts are expected to deliver insight without authority over inputs. Engineers are asked to industrialise use cases that rely on informal workarounds. The result is latent fragility. In other words, systems  appear to function well until demand spikes or scrutiny intensifies.

What differentiates progress is not industry or geography, but leadership posture. Where boards and CEOs treat data as infrastructure—akin to finance or risk—momentum builds. Budget follows clarity, and ownership becomes explicit. Where initiatives are driven from the middle, they struggle against inertia, competing incentives and cultural fatigue. Data transformation, it turns out, is less about persuasion than sponsorship, some leaders admitted.

Closing thoughts 

The emerging consensus is sobering but constructive. AI will not rescue poor data practices but will rather expose them. Organisations that succeed will be those willing to confront uncomfortable truths early, such as maturity is uneven, ownership is political, and governance is a prerequisite, not a brake. 

In an era of accelerating technological possibility, strategic advantage may belong not to the most ambitious adopters, but to the most honest ones, as we have found in previous debates on the same themes.

This Food for Thought was made in partnership with Confluent