Sanitized portfolio replica of a production dashboard — synthetic data, real method. The production version adds four more tabs, campaign-level tables, and click-lineage verification. Read the case study.
Paid-Search Funnel Decomposition
Is paid-search conversion really falling — or is spend shifting into tactics that never converted well? An exact decomposition splits the revenue change into eight additive effects, so mix shift can't masquerade as funnel failure. Every number recomputes client-side for any pair of ranges.
Compare any two spans
Data is weekly grain; selections aggregate exactly. Primary is the span you're explaining; Comparison is the base it's measured against.
Primary — Focus
to
Comparison — Baseline
to
Why revenue moved — eight additive effects
Log-mean (LMDI) decomposition of ΔRevenue. The eight bars sum to the total change to machine precision — no residual, no "other".
Mix vs. rate, by tactic share of paid-search clicks
The Simpson's-paradox guard: blended conversion can fall while every tactic's own funnel holds.
Funnel rates by tactic
Efficiency
cost side · lower is better on costs
Synthetic weekly panel, deterministic generator — figures are illustrative; the decomposition identity, the controls pattern, and the table conventions mirror the production build. Identity: Revenue = Σtactic Clickstotal · share · (Sess/Click) · (Shop/Sess) · (Cart/Shop) · (CO/Cart) · (Orders/CO) · AOV.