Data products and data contracts: what they are and who owns them

Two definitions

Data product

A dataset, or a set of related ones, with a named owner, a clear purpose, documented meaning, and quality and availability targets that its users can rely on.

Data contract

An agreement between a data producer and its consumers covering structure, meaning, quality checks and service levels, written so that software can test it.

Producer

The team that creates and changes the data, usually close to the source system.

Consumer

Any team, report, model or agent that depends on the data.

The point of both is accountability. A shared table that anyone can change and nobody owns is the usual source of the "three versions of revenue" problem. In dbt Labs’ 2026 survey, ambiguous data ownership was a persistent obstacle for 41% of respondents.¹

Where adoption stands

BARC surveyed 308 respondents in November 2025 for a study sponsored by Actian. It found 69% using data products operationally, at least in specific areas, and 61% using data contracts.² The same study found that 85% of companies with data products established company-wide had three or more AI projects in production, against 25% of the others. That is a correlation, and companies organised enough to run data products are likely different in other ways too.²

BARC also observed that the data mesh narrative behind data products "has lost traction".² The practices have survived the label. In its broader 2026 Trend Monitor, data products and marketplaces ranked 15th of 20 topics.³

Gartner listed highly consumable data products among its top data and analytics trends for 2025, advising leaders to focus them on business-critical use cases.⁴ In dbt Labs’ 2026 survey, the share of respondents prioritising "shipping data products faster" rose from 50% to 71% in a year.¹

The open standards

The Bitol project, hosted by the LF AI & Data Foundation, maintains two specifications. The Open Data Contract Standard (ODCS) describes contracts, and the Open Data Product Standard (ODPS) describes products.

Bitol says the number of organisations adopting the two standards rose from 9 to 114 in fourteen months, based on its own study ending in May 2026.⁷ That figure comes from the project itself and has not been independently checked.

Gartner expects the next step to be automated. By 2030, it forecasts, half of organisations will use autonomous AI agents to turn governance policies and technical standards into machine-verifiable data contracts.⁹

Starting with one dataset

Pick a dataset that several teams depend on and that breaks often enough for people to have noticed. Name an owner on the producing side. Write down, in plain words first, what each consumer actually needs from it: which fields, how fresh, what level of completeness, and who to call when it fails.

Then turn the most important of those promises into checks that run automatically, and publish the result where consumers can see it. Using an open format such as ODCS from the start costs little and keeps your options open if you change tools later.

Questions people ask