Media Mix Modeling · MMM
Media Mix Modeling / Marketing Mix Modeling
What is Media Mix Modeling?
- Estimating each channel’s sales contribution from aggregate data
- Works without individual tracking, so it holds up in a cookieless world
- Correlation-based estimates that need experimental validation
Adopting MMM changes the kind of questions asked in budget meetings. The discussion shifts from 'which ad did this lead come from?' to 'how much of last quarter’s revenue did advertising create?' It also resolves the problem of each channel’s dashboard claiming its own conversions, which inflated totals to two or three times the actual number of contracts, because everything is reconciled against one aggregated time series. In exchange, allocation is debated against a single company-wide revenue line rather than per-channel scorecards, so sales and marketing first need to agree to accept this number as a shared benchmark.
Two transformations are key when building the model. One is adstock, which sets a decay rate for how many weeks the effect of this week’s ad spend lingers. The other is a saturation curve, which models the range where each additional won put into the same channel yields less. As a baseline, 104+ weeks of weekly data is the minimum, and if you want to include more than eight channels, three years of data is safer. It serves a different purpose from attribution, which follows individual paths. MMM answers questions about total budget and cross-channel allocation, while attribution answers questions about campaign-level operations, so treating them as substitutes erodes trust in both. In B2B, contracts take 6–9 months to close, so using pipeline generated instead of revenue as the dependent variable makes the lag easier to handle.
The most common misunderstanding in practice is reading the contribution share directly as ROI. An output like 'search ads contribute 32%' is an estimate for the spend range you have actually run so far; it doesn’t mean the same share holds if you double the budget. The model doesn’t know ranges it has never observed. Another problem arises when spending patterns look the same year after year. If you split spend in the same proportions every month, the coefficients across channels absorb each other and the values swing wildly. Cutting the budget because you trusted 'exhibition contribution of 0' in that state can lead to a sharp drop in new pipeline two quarters later, and this really happens. Each quarter, deliberately switch some regions off and on to add validation from outside the model.
John KimInstructor John Kim, author of B2B Marketing Design and CEO of Nanum Consulting
Junyoung NohConsumer and generational trend analyst & AI content marketing specialistIt began in the 1960s, when econometric regression techniques were first applied to marketing budget allocation, and large consumer goods companies have long used it to separate the effects of TV and print advertising. In the 2020s it has drawn renewed attention because of privacy regulation and cookie restrictions.
Company A, an industrial sensor manufacturer with 210 employees and KRW 48 billion (≈ US$35,912,000) in annual revenue, spends KRW 2.4 billion (≈ US$1,795,600) a year on marketing. The budget is split across search ads, trade shows, industry media, and online seminars. But when the conversions reported by each channel dashboard were added up, the total came to three times the actual number of new contracts. Ahead of next year’s budget planning, the CMO pointed out the problem: “With these numbers, I can’t tell where to cut.” (Hypothetical example)
- Manager Kim on the marketing planning team put the most recent three years, 156 weeks of data, into a single table: weekly pipeline generated from the CRM, spend by channel, the exhibition schedule, and a raw materials price index.
- When an outside analyst ran the model, it showed that exhibition impact peaked four weeks after spend, and that once search ads passed KRW 60 million (≈ US$44,900) a month, there was a range where each additional won delivered half the effect.
- When the head of sales said, 'If exhibitions contribute only 8%, let's cut them next year,' Manager Kim pushed back. The figure was underestimated, she said, because of the 7-month lag before a contract closes, and she asked to re-estimate the model with a lag term.
- To validate, they switched search ads off completely for eight weeks in three sales territories in the Yeongnam region while leaving the Seoul metropolitan area unchanged. The actual drop in inquiries came within 11% of the model’s prediction, so they accepted the estimate.
- After shifting KRW 400 million (≈ US$299,300) of a KRW 2.4 billion (≈ US$1,795,600) budget from search ads to trade shows and industry media, new pipeline generated the next year rose 18% and cost per lead fell from KRW 340,000 (≈ US$255) to KRW 270,000 (≈ US$200).