Data quality is the project, not the preparation for it

A pattern we see often: an organisation buys a capable BI tool, builds forty dashboards in three months, and quietly stops using them by month nine. The tool was fine. Nobody trusted the numbers.

Trust is a property of the pipeline

By the time a figure reaches a dashboard it has passed through extraction, transformation, joins and aggregation, any of which can be silently wrong. If a stakeholder cannot trace a number back to its source, disagreement has no resolution mechanism and the dashboard loses every argument it is part of.

Four practices that change the outcome

Test data like code

Assert row counts, uniqueness, referential integrity and value ranges on every load. A pipeline that fails loudly is far better than one that produces a plausible wrong answer.

Define metrics once

"Active customer" must have exactly one definition, versioned, documented, and referenced by every report. Most reporting disputes are definitional, not numerical.

Publish lineage

Every figure should be traceable to its source system and transformation. This single capability ends more disputes than any amount of dashboard design.

Show the caveats

If a metric excludes a region or lags by a day, say so on the chart. Hidden caveats destroy trust the first time somebody discovers one.

Spend the first third of any analytics programme on data quality and lineage. It feels slow, and it is the reason the dashboards are still open in year three.