A data contract is an explicit agreement between the supplier and the consumer of data: what does this data mean, what shape does it have, what can you expect from it, and who is responsible when it does not hold.
The word "contract" sounds heavy, but the idea is plain. Today, data flows run on implicit assumptions. Someone builds a report on an export and assumes column G always holds net revenue. Six months later a colleague changes that logic in the source system, with the best intentions. The report keeps working and is quietly wrong. A data contract makes assumptions like that visible and enforceable.
What it contains
The definition. What is this, in the language of the organization? "Order complete" means: physically shipped and invoiced. Not something else.
The shape. Which fields, which types, which units, which key is unique. This part is technical, which is exactly why it is easy to check.
The expectation. How fresh the data is, how complete, which values are allowed, what deviation is acceptable. This is the quality promise.
The owner. One person or team that set the contract and announces a change before it goes live.
Definition, shape, expectation and owner on one page.
Why it matters
Data contracts solve a problem organizations usually experience as human conflict: two departments with two numbers for the same thing. Without a contract that is an argument about who is right. With a contract it is an argument about which definition applies, and that one can be closed, recorded and reused.
For automation and AI it is even more direct. A system acting on data has to know when that data can no longer be trusted. A contract gives it a measurable point: if the promise breaks, the action stops, instead of quietly producing wrong decisions.
If the promise breaks the action stops, instead of failing quietly.
How heavy should it be
Far lighter than it sounds. One page per dataset is enough to start, and your first five contracts cover the five sources you actually lean on. A contract is not a governance programme; it is the shortest way to write down an assumption before someone breaks it by accident.
Data contracts are part of the meaning layer under an organization. How that layer relates to the rest is in The company brain: where your organization's knowledge lives.
A contract in four lines
A workable contract fits on one page. For an order-line dataset it looks roughly like this: the definition ("order complete = shipped and invoiced"), the shape (order number is the unique key, amounts in euros excluding VAT), the expectation (refreshed daily before 07:00, at most 0.5% missing rows) and the owner (the logistics manager, changes announced two weeks ahead).
