Enterprise AI ambitions are accelerating faster than many organisations’ data foundations can support them.
US CIOs and data leaders are increasingly discovering that the biggest obstacle to enterprise AI adoption is not access to models. It is whether those models can rely on trusted, governed, well-defined data.
Recent private US enterprise IT roundtables repeatedly returned to the same issues: inconsistent definitions, fragmented data ownership, poor data quality, uncertain data lineage, weak governance and difficulty determining which sources AI systems should be allowed to trust.
Participants discussed standardising definitions, creating confidence levels for trusted data, strengthening cross-organisational governance and improving structured data maturity before scaling AI.
For technology vendors, this creates a major commercial signal.
The next phase of enterprise AI will not be won by organisations with the most AI tools.
It will be won by organisations whose data can support reliable AI decisions.
And for vendors selling data platforms, governance solutions, analytics, observability, integration, cloud infrastructure and AI technology, data quality is rapidly becoming part of the AI business case.
Poor data quality is becoming an enterprise AI problem
Generative AI can make information easier to access.
It cannot automatically make incorrect information correct.
That distinction is critical.
When an employee asks an AI assistant a business question, the answer may be based on:
CRM records.
Financial systems.
Product databases.
Documents.
Data warehouses.
Emails.
Customer records.
Operational systems.
Unstructured files.
Third-party datasets.
If those sources conflict, contain duplicated records or use different definitions, the AI inherits the problem.
Several US IT leaders specifically raised the risk of AI hallucinations caused not simply by model behaviour, but by feeding AI systems incorrect or poorly governed information. Participants emphasised establishing data-quality practices and defining authorised data sources before expanding enterprise AI usage.
This is an important distinction for vendors.
Many AI pitches focus on the intelligence layer.
Enterprise buyers increasingly need help with the trust layer beneath it.
AI readiness is becoming data readiness
Enterprise leaders may talk about becoming “AI ready”.
But much of that readiness is actually data readiness.
Can the organisation identify its authoritative data sources?
Does everyone agree on basic business definitions?
Can data be accessed securely?
Is sensitive information classified?
Can the organisation trace where information originated?
Are duplicate records controlled?
Can data quality be measured?
Can AI systems distinguish trusted sources from low-confidence ones?
If the answer to several of these questions is no, enterprise AI scaling becomes far more difficult.
US IT discussions included approaches such as creating trust lists and confidence levels for different data sources, improving structured data environments and defining clearer data-quality requirements before models are given broader access.
This shifts the commercial opportunity.
A data-governance project may no longer be justified only as:
“Improving data management.”
It can increasingly be positioned as:
“Creating the trusted data foundation required for enterprise AI.”
That is a much stronger boardroom narrative.
The enterprise AI stack is only as trustworthy as the data beneath it
The AI market often focuses attention on models.
Which LLM?
Which copilot?
Which agent platform?
Which AI provider?
Those are important decisions.
But the model sits on top of a much larger enterprise data environment.
A simplified view looks like this:
| Enterprise AI layer | Key buyer question | Vendor opportunity |
|---|---|---|
| AI models and applications | What can AI do for the business? | Generative AI, copilots, agents and automation |
| Governance and controls | What is AI allowed to do? | AI governance, policy and risk management |
| Data access | Which information can AI use? | Identity, security and access management |
| Data trust | Which information should AI believe? | Data quality, observability and lineage |
| Data architecture | Where does information live? | Cloud data platforms, warehouses, lakes and fabrics |
| Business definitions | Does everyone mean the same thing? | Semantic layers, catalogues and metadata |
| Source systems | Where does the data originate? | ERP, CRM, operational and industry platforms |
AI applications may be the most visible layer.
But failures lower in the stack can undermine everything above them.
For vendors, this means AI conversations increasingly create adjacent buying opportunities across the entire data ecosystem.
Enterprise data governance is moving closer to the boardroom
Data governance has historically struggled with perception.
It can sound administrative.
Policies.
Ownership frameworks.
Data stewards.
Catalogues.
Definitions.
Classification.
Documentation.
These things rarely generate the excitement of a generative AI demonstration.
But AI is changing the strategic relevance of governance.
When an AI system starts using enterprise data to recommend actions, generate analysis or automate workflows, basic governance questions suddenly become business-critical.
Who owns this data?
Is it accurate?
Can this user access it?
Can this AI agent access it?
Can the output be trusted?
What happens if two systems disagree?
Who is accountable if an automated decision is based on poor information?
US enterprise leaders discussed cross-organisational governance frameworks, data classification, privacy controls and the growing need to involve multiple functions as AI becomes more autonomous.
That moves data governance from the back office into the AI strategy conversation.
Common definitions are becoming essential for trustworthy AI
One of the least glamorous data problems is also one of the most damaging.
Organisations often do not agree on what their own words mean.
What qualifies as a customer?
What is an active account?
What is churn?
What is revenue?
What counts as urgent?
What is a qualified opportunity?
What does “high risk” mean?
Different departments can use the same term while calculating it differently.
In the roundtables, leaders described inconsistent metrics caused by different definitions across clinics and business units. Participants highlighted the need for a common business language and cleaner governance so analytics could produce consistent results.
The problem becomes more serious with AI.
A human analyst may notice that two reports define a metric differently.
An AI system may simply process both.
That makes semantic consistency increasingly important.
This creates opportunity for vendors offering:
Data catalogues.
Semantic layers.
Master data management.
Metadata management.
Business glossaries.
Knowledge graphs.
Data marketplaces.
The value proposition is no longer just easier analytics.
It is making enterprise information interpretable by both humans and AI.
Data silos are now blocking more than reporting
Data silos are not a new problem.
But AI amplifies their cost.
Traditionally, fragmented data meant:
Slow reporting.
Manual reconciliation.
Duplicate analysis.
Inconsistent dashboards.
Today, the same fragmentation can limit what AI systems are able to understand.
Enterprise leaders discussed both technical and political data silos, noting that governance and organisational barriers can be more difficult to solve than technology itself. They also highlighted the disconnect that can exist between technically available data and whether business users can actually access and understand it.
For vendors, this is an important warning.
Do not assume the problem is simply moving all data into one platform.
The buyer may also be dealing with:
Ownership disputes.
Departmental incentives.
Security concerns.
Different data maturity levels.
Compliance restrictions.
Poor business definitions.
Low user trust.
A stronger vendor proposition therefore combines architecture with governance and adoption.
Trusted data does not mean unrestricted data
One of the tensions enterprise buyers face is making data more accessible without making it uncontrolled.
AI creates pressure for broader access.
Employees want copilots to answer questions across internal information.
Analysts want easier access to datasets.
AI agents may need to interact with multiple systems.
But not every user or AI process should have access to every dataset.
Roundtable participants discussed tiered access models where users received different levels of data access based on expertise and governance requirements. Other discussions focused on data classification, privacy, ethical use and protection as foundational requirements for AI.
The commercial opportunity for vendors is therefore not simply “democratise all data”.
It is:
Make trusted data easier to use within appropriate controls.
That opens buying conversations around:
Identity.
Role-based access.
Policy enforcement.
Data masking.
Classification.
Data loss prevention.
Privacy.
Consent.
Auditability.
AI access governance.
Data observability is becoming part of AI reliability
AI increases the importance of knowing not just what data exists, but whether it is healthy.
Where did the data come from?
When was it updated?
Has the pipeline failed?
Has the schema changed?
Is the value complete?
Is the dataset stale?
Has a source definition changed?
Can the output be traced?
Enterprise leaders highlighted data quality and observability as critical components of AI implementation, with some organisations using existing data-lake investments and AI-generated metadata to address information gaps.
That creates a strong opportunity for data-observability vendors.
Historically, observability may have been pitched mainly around pipeline uptime and engineering efficiency.
In an AI environment, the value proposition expands.
Data observability becomes AI reliability infrastructure.
If an AI system depends on enterprise data, failures upstream can become incorrect outputs downstream.
That makes monitoring data health increasingly strategic.
Generative AI may actually increase demand for structured data
There is a popular assumption that powerful LLMs will eventually make structured enterprise data less important.
Why clean and structure everything if AI can understand unstructured information?
The roundtables suggest enterprise leaders are not convinced it is that simple.
Participants debated structured versus unstructured data and the future role of LLMs, while still emphasising business context, clear objectives and data-quality requirements. Some organisations were actively continuing their journey towards more mature structured-data environments before deeper LLM integration.
This is commercially significant.
AI may make unstructured data more usable.
It does not eliminate the need for:
Reliable identifiers.
Consistent relationships.
Business definitions.
Transaction integrity.
Metadata.
Data lineage.
High-confidence reference data.
For critical enterprise processes, structured data may become more valuable precisely because AI needs reliable anchors.
Agentic AI raises the stakes again
The risk becomes greater as enterprises move from generative AI towards agentic AI.
A chatbot answers.
An agent acts.
That difference is fundamental.
An agent may:
Update records.
Approve actions.
Trigger workflows.
Generate communications.
Move data.
Interact with other systems.
Perform compliance checks.
The more autonomous the system becomes, the more important its underlying data quality becomes.
Roundtable participants discussed using agentic AI for data cleansing and compliance-related automation while also raising concerns around accountability, risk management, privacy and adversarial testing.
For vendors, this creates a useful rule:
The more autonomous the AI, the higher the data-governance requirement.
A hallucinated answer is problematic.
An automated action based on incorrect data can be significantly worse.
What US enterprise IT buyers now need from data vendors
The data market is crowded.
Warehouses.
Lakehouses.
Fabrics.
Catalogues.
Governance platforms.
Observability.
Integration.
Master data management.
Analytics.
AI data layers.
The buying question is increasingly not:
“Which technology category do you fit into?”
It is:
“How does your solution help us create trusted data for AI and business decisions?”
The strongest vendor propositions should address several buyer priorities.
| Buyer priority | What vendors should demonstrate |
|---|---|
| Data quality | How inaccurate, incomplete or inconsistent data is identified |
| Data trust | How authoritative sources and confidence levels are established |
| Governance | How ownership, policies and accountability are managed |
| Data access | How users and AI receive appropriate access |
| Standardisation | How common definitions are maintained across the enterprise |
| Lineage | How data and AI outputs can be traced back to sources |
| Observability | How data failures are detected before they affect decisions |
| AI readiness | How the data foundation supports copilots, LLMs and agents |
| Business usability | How non-technical users can find and understand trusted data |
This is where vendors can differentiate.
Not through another abstract AI promise.
Through practical trust.
What vendors should change in their messaging
Many enterprise data vendors still sell primarily around architecture.
That language matters to technical buyers.
But the broader AI business case creates a stronger commercial narrative.
| Traditional vendor message | Stronger enterprise AI message |
|---|---|
| “Modernise your data architecture” | “Build the trusted data foundation your enterprise AI strategy depends on” |
| “Improve data quality” | “Prevent unreliable data from becoming unreliable AI decisions” |
| “Break down data silos” | “Give AI and business users governed access to trusted enterprise information” |
| “Implement a data catalogue” | “Make trusted enterprise data discoverable and understandable by people and AI” |
| “Improve data observability” | “Detect upstream data failures before they compromise AI outputs” |
| “Build a data fabric” | “Connect distributed data while maintaining governance, classification and control” |
| “Automate data cleansing” | “Improve AI readiness by continuously strengthening the quality of underlying data” |
The difference is positioning.
Data technology becomes easier to fund when it is connected to a strategic priority the board already cares about.
Right now, AI is one of those priorities.
AI vendors also need to take data quality seriously
The message is not only relevant to data vendors.
AI solution providers need to understand it too.
A vendor may have an excellent model.
But enterprise results will depend heavily on customer data.
That means AI vendors should be prepared to answer:
What data quality is required?
How are trusted sources identified?
How does the system handle conflicting information?
Can outputs show source provenance?
How is low-confidence data treated?
What happens when upstream information changes?
How can enterprise teams audit the result?
Does the platform support authorised data boundaries?
The vendor that openly addresses data dependency may appear more credible than one promising AI magic regardless of the customer’s environment.
Data quality is becoming part of AI ROI
Poor data does not only create technical risk.
It damages the business case.
If employees cannot trust AI outputs, adoption falls.
If analysts must verify every answer manually, productivity gains shrink.
If incorrect information triggers rework, efficiency disappears.
If governance failures delay deployment, time to value increases.
If unreliable data causes business errors, risk rises.
This means data quality is directly connected to AI ROI.
The roundtables reflected this broader connection between data governance, AI value and scaling. Enterprises are looking for measurable business outcomes while simultaneously strengthening the data foundations required to support them.
For vendors, this creates a powerful commercial argument:
Better data is not merely a technical improvement. It protects the return on AI investment.
The strongest enterprise data strategies combine technology and culture
Data quality is not solved entirely by software.
Many data problems originate in behaviour.
People enter data differently.
Departments create local definitions.
Teams keep private spreadsheets.
Ownership remains unclear.
Users distrust central systems.
Governance becomes something IT “does to” the business.
Recent US discussions on data-driven culture highlighted exactly these challenges, including technical and political silos, inconsistent business language, training needs and the importance of change management.
This matters for vendors.
A data platform may technically solve the problem.
Enterprise adoption determines whether it actually does.
Strong vendors should therefore help buyers think about:
Ownership.
Stewardship.
Training.
Adoption.
Business curation.
Executive sponsorship.
Governance operating models.
User experience.
The product may be technical.
The transformation is organisational.
Why vendors need to reach enterprise buyers before the data strategy is fixed
Data architecture decisions have long consequences.
Once an enterprise establishes:
Its governance model.
Its AI architecture.
Its trusted data sources.
Its access policies.
Its cloud platform.
Its semantic layer.
Its data marketplace.
Its preferred integration patterns.
The vendor landscape narrows quickly.
This is why early buyer access matters.
By the time a formal RFP appears, strategic assumptions may already be established.
The Leadership Board helps technology solution providers engage senior US CIOs, CTOs, data leaders and enterprise IT decision-makers around active priorities before the buying conversation becomes purely procurement-led.
These conversations help vendors understand whether the buyer’s immediate challenge is:
Data quality.
Governance.
AI readiness.
Data access.
Observability.
Cloud modernisation.
Analytics.
Or a combination of them.
That context is what allows a vendor to position against the real enterprise priority rather than simply presenting another product.
Meet US enterprise IT buyers with active data and AI priorities
Data quality may be the least glamorous AI priority and one of the most valuable
The enterprise AI conversation is evolving.
Models will improve.
AI agents will become more capable.
Automation will increase.
But greater AI capability does not reduce the importance of trustworthy data.
It increases it.
The most important signal from recent US enterprise discussions is clear.
Organisations are trying to establish:
Trusted data sources.
Consistent definitions.
Stronger governance.
Better quality controls.
Appropriate access.
Clearer accountability.
More mature data foundations.
Because the more an enterprise relies on AI, the greater the cost of feeding it information that cannot be trusted.
For vendors, that creates a major commercial opportunity.
Do not sell data quality as housekeeping.
Do not sell governance as bureaucracy.
Do not sell architecture as technology for technology’s sake.
Position the data foundation for what it increasingly is:
The infrastructure that determines whether enterprise AI can be trusted, scaled and turned into business value.