Can a 200-year-old advertiser find a quarter of its digital spend doing nothing?

Can a 200-year-old advertiser find a quarter of its digital spend doing nothing?

The question for any brand with a nine-figure budget split across dozens of channels: which slices are actually moving sales, and which just coincide with sales?

The method. Procter & Gamble, around 2017, applied media mix modeling — a regression-based, top-down approach that relates aggregate sales over time to spend across channels, controlling for price, seasonality and distribution. Unlike click-based attribution, it needs no user-level tracking; it reads the macro relationship between dollars in and sales out.

What they reported. P&G publicly cut over $100M in digital ad spend (one widely cited figure was ~$200M across a period) after the modeling and follow-on tests showed large portions delivered no measurable sales lift — partly from bot traffic, partly from ads served where they couldn't work. The reported outcome: reach and sales held steady despite the cut.

The nuance. Media mix modeling is correlational at its core — it can mistake a channel that rises alongside sales for one that causes them. P&G's strength was pairing the model with validation experiments, so the cuts rested on tested causal claims, not regression coefficients alone.

Bottom line for practitioners: media mix modeling is the right lens for budgets too fragmented or privacy-constrained for user tracking, but its coefficients are hypotheses. Calibrate the model against at least one holdout experiment before you move money. P&G's headline cut survived because the regression was checked against reality, not trusted on its own.