Eight years of evidence, five ways to spend the year
A global tea business planning its US media year against a decade of mix-model history.
The situation
A two-billion-dollar global tea business planning its US media investment for the coming year, with a portfolio of brands and sub-brands, a large ready-to-drink partner whose advertising created a halo the company did not pay for, and a national reach target that linear television alone could no longer meet efficiently. Marketing mix models had been run every few years for a decade, by different vendors, with results filed and rarely re-read.
What I did
Pulled every model result since 2015 into one table: channel, creative, consumer segment, return, and any halo the creative had thrown onto adjacent lines. That table became the evidence base for an expected return by channel, stated as a range rather than a point. On top of it I built a monthly plan for each brand with spend by channel, expected return, expected lift, impressions, and estimated unique reach against the brand's goal audience, so the plan could be checked for reach sufficiency and for return at the same time. Then five versions of the year: the base, a masterbrand-heavy version, a masterbrand-plus-iced version, a version that only funded channels with a demonstrated positive return, and a sub-brand version.
What changed
The recommendation shifted the video mix from a television-first plan toward a blended plan with streaming and online video carrying most of the reach, at materially lower cost for equal or greater unique reach. More usefully, the leadership team could see what each way of spending the year was likely to return before committing to any of them, and could see why.
What I would do differently now
The workbook was manual and the return assumptions lived in my head as much as in the cells. It was the first version of what later became GrowthOS: the same logic, with the evidence base kept current, the scenarios re-run each period, and the confidence carried on every figure.