The Goal
The proximate trigger: in Q1 2026, paid media spend was up 28% YoY while revenue was down 21% YoY — the sharpest divergence the account had seen in three years.
The bigger context underneath it: account ROI had been deteriorating across the full three-year window (9.84x → 5.09x), and one specific metric — Average Order Value — had collapsed 35% in fifteen months ($747 → $486), the most precise discontinuity in the account's history.
The brief: diagnose what was actually driving the deterioration, separate measurement-level issues from real demand shifts, and build a ranked, evidence-based case for what specifically to change next.
Three sub-questions to disentangle:
· Is the recent inflection about unit economics, paid media inefficiency, or measurement? · Where in the channel mix is the inefficiency concentrated? · What specific levers, ranked by expected impact, should we pull?
How We Went About It
Pulled 40 months of paid media spend across Google, Meta, and Bing — joined with order-level revenue (ERP) and GA4 traffic at the month level.
Ran correlation analyses on spend → revenue, including lagged variants (T+1, T+2, T+3 months) to test for delayed effects. Segmented Brand vs Non-Brand spend to test the cannibalization hypothesis. Sliced order composition data to test for cart-value or measurement breaks.
The whole thing got built as a single self-contained HTML file — Chart.js + Tailwind via CDN, twelve independently navigable tabs, every chart traceable back to the underlying data. The format was a deliberate choice: the client should be able to open the file offline, audit any claim, share it via email, or revisit any single finding without needing the rest of the analysis.
Key Insights
Two findings carried the weight.
(1) Heavy Brand-campaign spend was cannibalizing organic. The Brand traffic share rose 7–10 percentage points without total non-CPC traffic moving — the textbook fingerprint. Most people typing the brand name into Google were already going to land on the site through organic, direct, or email. Bidding aggressively on those queries moved revenue from "free" channels to "paid Brand" without lifting incremental revenue. Recommendation: shift toward a 70/30 Non-Brand/Brand mix and let Non-Brand do the brand-development work.
(2) Cart-value mechanics had broken. Single-SKU orders only, exactly 1.0 items per order — the most precise product-mix discontinuity in the account's history. Either a bundling change at the site level or a measurement issue, but either way: months with higher add-on values per order had consistently been the best months. Reactivate cross-sell on PDPs and post-add-to-cart upsell modules; clean up the data with the client team in parallel.
Plus a meta-finding worth naming: the deliverable format matters as much as the analysis. A single auditable HTML file changes how a client engages with the work — they read more of it, return to it, and reference it in their own decision making. Decks get filed; auditable analyses get used.