The Problem
A national consumer brand ran roughly $100 million of television over a single four-month campaign, the stretch that decides its year. The money bought a rotation of ten to fifteen commercials, fifteen and thirty seconds each. The decision was where every dollar should go: which markets, which spots, and which weeks. The objective was customers through the door, not awareness for its own sake, and four months was too short to waste a week.
A single national plan set in advance could not get this right, because two forces move the result and neither holds still. Markets differ. Response to advertising in any market depends on its consumers, their incomes, and how much weight is already on the air there. And a commercial is not static. It takes time to build before it works, then wears out as people see it too often, until added airtime returns less and less. The decision had to be made again each week, against what the market was actually doing.
The Results
The work produced a weekly allocation plan — market by market, which spots to keep on the air, which to retire, which new ones to introduce, and how much weight to put behind each. On the same budget, it raised the return by an estimated $10 to $15 million a year.
Three reads fed every recommendation. A market read estimated how customer response in each market moved with advertising weight, given local demographics. An analytics read modeled build and wear-out, so the plan could tell a spot that was still gaining from one that had stopped paying. And a primary-research read kept the models honest. A survey firm polled viewers across the country every few days on which commercials they recalled and liked, and a Kalman filter folded those readings back in each cycle, so the plan corrected against what audiences were doing rather than what the model had assumed. Where the three agreed, the dollars moved.
The approach outlasted the first campaign and the original engagement. After it had proven out, the work was picked up again under a later engagement to keep it running, a sign the lift was real and repeatable. The point was never a marketing-mix model on its own. It was reading the whole system behind the result, market by market and week by week, and turning that into a plan the media buy could act on by Monday.
Key Techniques
- System-dynamics modeling of advertising build and wear-out, so airweight matched where each spot was in its life.
- Market-level response estimation: customer response to gross rating points by market, conditioned on local demographics.
- A Kalman filter that folded near-real-time survey readings into the model each cycle, correcting the plan against observed recall and favorability.
- Continuous national telephone surveys measuring ad recall and liking every few days, as an independent read on the model.
- A weekly operating cadence, issuing rotation, copy, and market-placement decisions on a schedule the media buy could execute.