Data problems usually look like people problems

When a report is wrong, the first response is often to ask who owns the spreadsheet. The deeper issue is usually that the same concept is defined differently across systems, or that the data has no clear path from source to decision.

Good data foundations do not remove every disagreement. They make the disagreement visible. A shared definition, a known source and a clear owner turn a recurring argument into a one-time decision.

That is why data work is rarely just a technical project. It changes how teams agree on what they are measuring.

Start with the decisions that matter

It is tempting to build a warehouse for every dataset the business might eventually need. A better starting point is a small number of decisions that are currently slow, contested or unreliable.

List those decisions and work backwards. What does the decision-maker need to see? Where does that information originate? Who is allowed to change it? What would make the number trustworthy?

This approach keeps the foundation focused. It also creates visible value early, which makes it easier to sustain the less visible work that follows.

Treat definitions as part of the product

A metric without a definition is an invitation to disagree. Revenue, active customer, churn and utilisation can all mean different things depending on the team and the system.

Write the definitions down, assign owners and connect them to the implementation. When the definition changes, the change should be visible to the people who use the reporting.

This is unglamorous work, but it is often the difference between a dashboard that creates confidence and one that creates another meeting.

Build reliable paths from source to use

A data foundation needs a dependable route from operational systems to analytics and applications. That route should include validation, monitoring and a clear response when something breaks.

It does not need to be complicated at the start. A small number of well-understood pipelines with tests and ownership are more valuable than a broad integration layer nobody trusts.

As the business grows, the same principles scale: clear sources, explicit transformations, quality checks and visibility into freshness and failure.

  • Document the source of each important metric.
  • Test the transformations that matter to decisions.
  • Monitor freshness and failure visibly.
  • Give every critical dataset an owner.

Make access intentional

Access is part of data quality. If people cannot find the right data, they will build a local copy. If too many people can change sensitive data, trust erodes.

Define roles around the work people do. Provide a curated path for common questions and a controlled path for deeper analysis. Keep sensitive information out of systems that do not need it.

Good access design reduces both shadow reporting and avoidable risk.

Build in increments, but keep the whole picture

A data foundation is never finished. The practical approach is to deliver it in increments while keeping a clear model of the whole.

Start with the decisions and datasets that create the most value. Establish the patterns, definitions and ownership that later work can reuse. Then expand deliberately rather than accumulating disconnected pipelines.

The result is not a perfect data platform. It is a business that can answer important questions faster, with less reconciliation and more confidence in the answer.