The Maze: Discounts look simple until customers learn the schedule. The Magic Numbers/WARC MMM guide shows a brand that moved from modest -3% to -5% promotions into deeper -8% to -10% cuts. The old pattern was predictable. Deeper discounts lifted revenue. Then Year 3 broke the curve. The brand spent more margin and got a weaker response. That is not a media problem. That is a customer-training problem.
The promotion curve changed before the budget owner noticed. The source example shows blue Year 1-2 promotions forming a clean downward relationship: as discount depth increased, revenue response rose in a fairly predictable way. The orange Year 3 points sit in the deeper discount zone, near -8% to -10%, but below the old response line. The useful lesson is blunt: a deeper deal is not automatically a better deal. Once buyers expect the markdown, the discount becomes a reservation price with a nicer suit.
MMM gets more interesting when it leaves the media silo. The guide says cost-benefit analysis is a core MMM output, but it applies beyond channels. Page 20 uses price promotions to make that point. Each promotion is a dot, discount depth sits on the x-axis, and revenue response sits on the y-axis. The recommendation was to revert to the old strategy. That is the kind of answer media-only measurement cannot give, because the damage sits in pricing power, not CPM.
The operating question is whether the model sees the full commercial system. The same guide tells marketers to ask suppliers whether models include price, economy, seasonality, COVID, and how those factors affect the business. That sounds basic. It is not. A retailer can grow sales through advertising, promotions, availability, assortment, service, and macro luck. If the model only reads ad spend, every answer leans toward media optimization. Useful, but incomplete. Sometimes the smartest media move is to stop teaching customers to wait.
Automation helps only if the inputs are not lazy. Google is bringing Meridian into Google Analytics 360 so teams can unify first-party, cross-channel data, measure causal performance, and run predictive scenarios. That is good plumbing. But better plumbing does not fix a narrow question. Promotional depth, price elasticity, and customer satisfaction still need to enter the model. Otherwise the machine gets very good at optimizing the wrong half of the business.
Why it matters: Retailers are not short of dashboards. They are short of honest measurement boundaries. Promotions can create short-term volume and long-term dependency in the same quarter. The source example is small, but the trap is large: if a customer learns that patience is rewarded, future demand comes with a coupon attached. MMM should not just defend media spend. It should show when the commercial model is eating itself.
