Why Promotions Matter
Promotions (discounts, sales, free shipping events, BOGO offers) produce large lift in conversion that is not driven by media. If the MMM does not model promotional intensity, it attributes the lift to whatever media was running during the promotion. Black Friday paid spend looks 5x more effective than it is because the model conflates promotional lift with paid effect.
Step 1: Build the Promotional Intensity Series
Weekly promotional intensity as a quantitative variable. Options: average discount depth across active SKUs, share of revenue from discounted SKUs, promotional spend as a percentage of revenue. Pick one approach and stick with it; the choice matters less than the consistency.
Step 2: Add Promotional Calendar Indicators
Specific high-promotional events (Black Friday, Cyber Monday, brand-specific anniversary sales, Prime Day for Amazon-dependent brands) get binary indicators in addition to the intensity variable. The events have large one-week spikes that the smooth intensity series does not capture.
Step 3: Model the Pull-Forward Effect
Promotions pull demand forward from future periods. The week after a major sale typically has lower-than-baseline demand because buyers who would have purchased in that week purchased during the promotion. Model this as a lagged effect of the promotional intensity, or use a negative-effect indicator for the post-promotion week.
Step 4: Distinguish Brand-Driven From Retailer-Driven Promotions
For brands selling through retail, retailer-driven promotions (the retailer's sale events) produce different effects from brand-driven promotions (the brand's own sales). Model separately if both are material; the response curves and pull-forward dynamics differ.
Step 5: Validate Decomposition
After fitting, inspect the promotional contribution decomposition. Promotional spend should attribute meaningful but not dominating share of revenue; if promotions are contributing 60+ percent of total revenue, the promotional variable is over-fit. If essentially zero in a brand with material promotional activity, it is under-fit.
How Presenc AI Helps
Presenc AI provides AI visibility data that disentangles AI-driven lift from promotional lift. Without the AI variable, AI-driven demand can be absorbed into the promotional indicator (especially when promotions and AI campaigns run concurrently); with both variables in the spec, the model identifies their separate contributions.