Weekly reporting

The weekly promotion report drew on four systems: the store, email, shipping and web analytics. Its numbers were worked out from scratch each week, with no fixed method, so a comparison with last week could not be reproduced, and nothing stopped a confident conclusion from resting on thin evidence.

I built a pipeline that collects each week's data into one place and computes the week-over-week figures the same way, every time. Missing data is labeled missing rather than shown as zero, and every report is checked by a separate AI reviewer that had no part in writing it, before it is delivered.

The weekly report is ready sooner. Each week is measured the same way, so one week can be compared fairly with the next, and next week's promotion is planned from findings that have already been through that review. Every computed number can be rechecked from the saved data, and nothing missing is hidden.