Most reporting projects do not fail on charts. They fail the first Monday a store manager and a regional manager look at the same week and see two different numbers. From then on, every meeting starts with an argument about the data instead of a decision.
Three rules we build by
- One resolver. Every page asks the same function what “this week” or “season to date” means. No page computes its own dates.
- Both sides on one window. A ratio like conversion is only honest if sales and traffic stop at the same hour.
- Check against the source, every load. After each data load we reconcile our totals with the raw feed and with the client’s own BI, and flag any gap before anyone sees it.
None of this is glamorous. It is what makes the AI layer worth having: commentary is only as good as the numbers underneath it.