An analytical join that doubles revenue and an imperfect model score
A fictional business has one hundred distinct completed orders with revenue of ten thousand. An analyst joins the order table to a customer-history table that contains two matching rows for each customer and forgets the effective-date condition. The resulting report shows two hundred rows and revenue of twenty thousand. Separately, a supplied evaluation set contains one thousand cases, including one hundred actual positive cases. A model correctly flags ninety of those positives but also flags ninety negative cases. All figures are classroom data rather than actual customer records or a clinical decision system. The manager wants to announce doubled revenue and a model that is simply ninety-percent accurate without explaining the transformations or errors. The team must separate data-quality evidence from business growth and distinguish useful model measures.
1. Explain the revenue result and identify an appropriate validation.
Read the worked answer
The join duplicated each order twice, so the apparent growth is not evidence of additional sales. Validate the customer-history keys and effective dates, check distinct order counts, and reconcile the corrected total to the original 10,000.
2. Calculate precision and recall for the supplied model.
Read the worked answer
Precision is 90/(90 +90) =50%. Recall is 90/100 =90%. The same model can have high recall and many false positive flags; the decision requires error costs and context, not one unlabeled percentage.