Choose a comparison design
Where practical, define holdout or randomized groups before exposure. If experimentation is not available, use a documented comparison method and disclose the biases it cannot remove.
Measure the intended outcome
Acquisition, order value, product adoption, and repeat purchase require different outcome windows and customer populations. Do not substitute an easy metric for the stated objective.
Account for competing explanations
Pricing, media spend, inventory, seasonality, merchandising, customer selection, and concurrent promotions can move Shopify performance. Report these factors alongside the estimate.