The confident wrong number

How we work

Audit the data, agree the definitions

Reporting projects are usually specified as a visualisation problem and fail as a data problem. The records are duplicated, two systems disagree, or a field has been used to mean three different things as staff and processes changed. A dashboard built on that does not look doubtful — it looks authoritative, which is precisely why it causes worse decisions than no dashboard at all.

The second failure is definitional. Two departments reporting the same measure with different cut-off dates or inclusion rules will produce different totals forever, and the argument that follows gets blamed on the tool. Writing a definition for every measure and agreeing it before building is dull work that prevents most of it.

And automation is mostly about failure. Scheduling an extract is straightforward; deciding what happens when a source arrives late, a column changes type or a load half-finishes is the real work. A pipeline that fails loudly is far safer than one that quietly reports yesterday's figures as today's.