Unreliable data is the quiet killer of AI initiatives. A 2026 Drexel University-Precisely survey of 505 data and analytics leaders found that 43% identified data readiness as the top barrier to aligning AI with business objectives, ahead of infrastructure at 42% and skills at 41%. As organizations push AI into production, chief data officers are turning to data contracts to enforce quality controls before bad data reaches models and downstream systems.
A data contract defines the expectations between data producers and consumers, covering schema, ownership, availability, quality standards, and change management. These formal agreements are machine-readable, which allows them to integrate into automated governance environments. The goal is to establish expectations before data moves among teams, platforms, and AI systems.
"Organizations that can establish that level of consistency are much better positioned to build the trusted data foundation that enterprise AI depends on," said Anant Adya, executive vice president and head of Americas Delivery at IT services provider Infosys.
Accountability and compliance by design
Data contracts force organizations to delineate who owns and produces data, and who consumes it. That clarity matters as AI pulls data across domains and workflows, where unclear ownership makes it difficult to trace errors to their source.
"One of the shifts we're seeing is that AI is driving organizations to be much more deliberate about how data is owned, governed and shared across the enterprise," Adya said. "Those decisions increasingly influence the operating model just as much as the underlying technology."
Balazs Fejes, president and CEO of IT consultancy EPAM Systems, said data contracts make ownership explicit, but accountability should sit with data producers, not centralized governance groups.
Data contracts also support regulatory compliance. Jes Gonzalez, a principal consultant at ePlus, cited the EU's AI Act and ISO/IEC 42001 as frameworks with data governance provisions that data contracts can help satisfy. When paired with machine-readable metadata, contracts provide evidence of what happened to data during audits or investigations.
"It shows what happened when the data moved or changed, instead of forcing people to rebuild the story later from email, spreadsheets and guesswork," Gonzalez said. "That is what makes data contracts useful for compliance, audit support and AI governance. They turn policy into something that can be enforced and recorded in real time."
Applying governance and quality controls at the point where data is created, sometimes called shift-left or upstream governance, also preserves data team capacity. Without those controls, downstream teams absorb significant cleanup work, according to Gonzalez. "You can burn out your data team if you are not leveraging them well," he said.
Contracts are not a cure-all
Data contracts define structure, quality, and delivery requirements, but they do not explain business meaning. Fejes noted that a semantic layer provides the shared definitions, metrics, and relationships that AI agents need to interpret data consistently.
"Now that agents will start making decisions with minimal human interaction, the richness of the information required to support this data asset grows exponentially," he said.
Data contracts also require operational discipline. Stacking multiple contracts without strong standard operating procedures can create administrative and technical debt, Gonzalez said. Executive sponsorship matters for preserving funding and resolving cross-functional disputes, but operational accountability should remain with the teams that produce and consume data.
Adoption is gaining ground. A BARC survey of more than 300 organizations conducted in 2025 found that 69% had adopted data products, up from 48% the previous year. Among respondents, 61% reported using data contracts operationally in some areas. The report recommended treating AI systems as products, standardizing data contracts early, and correcting data quality issues at their source.
Where to start with data contracts
Organizations should determine whether data contracts address a material business or operational risk before committing to them. Gonzalez recommended starting with one to three use cases tied to business risks. A confidentiality, integrity, and availability (CIA) assessment can help identify sensitive and critical data, alongside factors like business impact, downstream dependencies, and recovery requirements.
"It's an opportunity for you to say, these are the datasets that, if impacted, have the potential to disrupt the business from a CIA perspective," Gonzalez said. Protected health information or payment card data could rank among the most critical assets. "Everybody states what type of data is important to them," he said. "Then you have the beginnings of an inventory for what could become a couple of solid data contracts."
For executives tracking AI strategy, data contracts are a governance mechanism with direct ROI implications. They reduce remediation work, support compliance audits, and clarify ownership before AI systems scale. The BARC data shows adoption is already widespread among data product programs, which means the practice is becoming table stakes rather than a differentiator. Leaders who treat data contracts as a discipline, not a document, position their AI investments on a foundation that holds.
Executives weighing next steps can start with a CIA assessment and one high-value use case. That approach builds momentum without overcommitting resources. For deeper context on how data governance fits into broader AI strategy, see AI for Executives & Strategy or the AI Learning Path for Chief Digital Officers.
Why this matters for executives and strategy leaders
Data contracts are an operational answer to a strategic problem: AI returns depend on data trust. The Drexel survey confirms data readiness is the top obstacle to AI alignment, ahead of infrastructure and skills. Executives who push accountability to data producers, pair contracts with semantic layers, and start with targeted use cases will spend less on cleanup and more on outcomes. Those who skip this step will discover the cost of bad data after AI systems are already in production, when fixes are most expensive.
Your membership also unlocks:
