Enterprise AI can look impressive right up until two people ask the same question and get two different answers.
That is where the conversation changes. What began as an AI capability discussion becomes a data trust problem. The model may be functioning. The interface may be intuitive. The pilot may have generated enthusiasm. But if Finance, Marketing, Operations and the AI assistant all interpret the same metric differently, the organisation does not have an AI problem. It has a data foundation problem.
Recent UK data-leader discussions convened by The Leadership Board exposed this tension repeatedly. Senior practitioners described inconsistent metric definitions, fragmented sources of truth, weak data ownership, poor metadata and the operational difficulty of moving AI from a promising proof of concept into a trusted production environment.
For data vendors, this matters because the buying conversation is moving underneath the technology. Enterprise buyers are increasingly asking whether the information beneath AI is consistent enough, governed enough and understandable enough for people to act on the output.
AI is exposing disagreements the organisation could previously hide
Traditional analytics allowed a surprising amount of ambiguity to survive. Different teams could maintain their own spreadsheets. Separate dashboards could contain slightly different definitions. A commercial team could calculate lifetime value one way while Finance used another. People often learned which report to trust through experience rather than through any formal semantic standard.
Generative AI makes that ambiguity much harder to ignore. The user is no longer navigating a known dashboard with familiar filters. They are asking an open question in natural language and expecting an authoritative answer.
One organisation described exactly this challenge. Different teams were using natural-language querying to create different versions of important commercial metrics. The appetite for AI was strong, but trust suffered because there was no sufficiently centralised semantic foundation behind the answers. The response was not simply to improve prompting. The organisation was building a semantic layer to standardise definitions before expanding natural-language access.
AI does not remove disagreement from enterprise data. It accelerates the moment when that disagreement becomes visible.
The semantic layer is becoming part of the AI control plane
The phrase “semantic layer” can sound technical and narrow. In an AI-enabled enterprise it is increasingly strategic. It is the mechanism that helps define what a metric means, which calculation is approved, what context applies and how different systems should interpret the same business concept.
Without that layer, an AI tool can be technically correct while still being commercially wrong. It may retrieve valid data from an approved system and still answer with a number that conflicts with the number another function uses to run the business.
This is why the market opportunity extends well beyond model providers. Vendors selling catalogues, metadata management, semantic modelling, lineage, observability, data quality and governance capabilities are increasingly sitting underneath the AI value proposition. The buyer may have entered the conversation looking for AI. The infrastructure that determines whether AI can be trusted is often much broader.
That reinforces a theme we have seen elsewhere in enterprise AI-ready data: buyers are becoming less interested in whether a model can generate an answer and more interested in whether the organisation can defend that answer.
Data quality has moved from hygiene to production risk
Poor data quality has always created waste. AI changes the scale and speed of that waste.
A dirty record inside a spreadsheet may inconvenience one analyst. A poorly governed definition exposed through an enterprise AI interface can influence hundreds or thousands of decisions. The issue is no longer only whether the source data is complete. Buyers need confidence in accuracy, freshness, permissions, lineage, calculation logic and context.
One roundtable example illustrated how substantial the underlying work can be. A large organisation described a two-year programme to create a unified data platform and semantic layer. The organisation reported a major improvement in data accuracy after implementation, while also stressing that the work remained ongoing. The commercially important point is not the individual percentage reported. It is that trusted AI foundations may require sustained data transformation rather than a fast model deployment.
Vendors that ignore that reality risk selling an AI acceleration story into an environment that cannot safely absorb it.
A model that can say “I don’t know” may be more valuable than one that always answers
Another important buyer signal is the growing value of uncertainty.
In regulated environments, confidently wrong answers are often more dangerous than incomplete answers. Senior leaders discussed guardrails that allow AI systems to acknowledge uncertainty, direct users toward approved sources and restrict responses when confidence is insufficient.
That is a subtle but important shift in what “good AI” means. Consumer AI often rewards fluency. Enterprise AI has to reward reliability. The best system is not necessarily the one that responds most naturally. It is the one that knows when the underlying data, permissions or context are not strong enough to support a decision.
For vendors, the proof point should therefore move beyond answer quality. Buyers need to understand how the product handles ambiguity, conflicting metrics, missing lineage and low-confidence situations.
The buyer is looking for an operating model, not another data tool
Technology alone cannot decide which definition of a business metric is correct. It can surface inconsistencies, enforce approved definitions and automate controls, but someone still has to own the decision.
That is why data ownership appeared repeatedly in the discussions. A common direction was centralised policy with decentralised accountability. Enterprise-wide standards need consistency, but business functions often need to own the meaning, quality and practical use of the data they understand best.
This creates a commercial requirement for vendors. Your solution has to fit the buyer’s operating model. It should clarify ownership rather than obscure it. It should make stewardship easier rather than creating another administrative burden. It should make it obvious when a definition changes, who approved the change and which downstream products are affected.
The same applies to metadata. As we have explored in our work on metadata problems that block enterprise AI, the commercial value of metadata is not documentation for its own sake. It is the context that allows people and machines to understand what they are actually using.
What enterprise buyers need to see
| Buyer concern | Weak vendor response | What stronger evidence looks like |
|---|---|---|
| Conflicting metrics | “Our AI can query all your data” | Approved definitions, semantic controls and conflict handling |
| Unclear ownership | “The platform centralises governance” | Named ownership, stewardship workflows and decision rights |
| Poor data quality | “AI will help clean the data” | Quality rules, issue visibility, thresholds and remediation evidence |
| Low-confidence outputs | “Our model is highly accurate” | Guardrails, uncertainty handling and approved-source signposting |
| AI scale | “We can deploy quickly” | Production controls, lineage, permissions and operational monitoring |
The sales pitch has to move below the model
The commercial mistake is to assume that AI excitement removes the need to discuss foundational data work. The opposite is happening. AI is making those foundations easier for executives to understand because failures are becoming visible in business-facing experiences.
A CFO does not need to understand lineage architecture to understand that two systems are producing different margin figures. A sales leader does not need to understand ontology design to recognise that the AI assistant is defining a customer differently from the CRM. A board does not need to understand metadata standards to know that it cannot govern a decision if nobody can explain where the number came from.
That gives data vendors a stronger commercial story. Do not position semantic consistency, governance, quality and metadata as housekeeping that has to be completed before the exciting AI work begins. Position them as the mechanisms that make enterprise AI commercially usable.
What data vendors should change
Sell trusted answers, not access to more data
Access is becoming easier. Trust is becoming harder. Show how your proposition creates a reliable path from source data to business answer.
Make semantic consistency visible in the demo
Do not demonstrate only the happy path. Show what happens when two definitions exist, when data is incomplete or when the user asks for a metric that has not been certified.
Connect governance directly to AI adoption
Governance should not appear as a separate compliance module at the end of the pitch. It should be part of the reason the buyer can safely expand access.
Help the buyer expose disagreement before AI does
Assessment, discovery and proof-of-value work should identify conflicting definitions, ownership gaps and weak source data early. Finding those issues is not a failure of the sales process. It is often the clearest route to a larger and more durable programme.
AI readiness is increasingly a data agreement problem
Enterprise organisations do not need every dataset to be perfect before they can use AI. They do need to know which data is trusted, what important terms mean, who owns the decisions and how the system behaves when those foundations are weak.
The vendors that understand this can move the conversation away from generic AI capability and toward the infrastructure that determines whether AI reaches production, earns user confidence and influences real decisions.
The Leadership Board gives technology vendors direct access to the priorities senior enterprise buyers are actively working through. If your proposition helps organisations create more trusted, governed and AI-ready data foundations, the strongest opportunity is often to engage while those requirements are still being defined.
Related reading
Explore Enterprise data buyer intelligence for the wider buying context.