Orders per session, conversion rate, click-through rate — a huge share of business metrics are ratios, not sums, and a ratio metric's variance can't be computed the way Chapter 2's t-test computed it. Averaging per-user ratios directly gets both the point estimate and the variance wrong. The delta method fixes both, and it's exactly as computable in SQL as a plain t-test.
Why "Average of Per-User Ratios" Is the Wrong Metric
Suppose the metric of interest is orders per session...