What Is Marketing Mix Modeling and Robyn?

Ads Manager tells you what Meta’s systems credit to your ads. Marketing mix modeling steps back and looks at your whole business over time: sales, every media channel, promotions and outside factors. It then estimates how much each one moved the result. Meta describes it as an econometric model built to help you decide how to split a marketing budget across channels, products and regions, and to forecast the effect of future campaigns.

How does marketing mix modeling work?

MMM relies on regression. Meta explains that it builds an equation describing a KPI, such as sales volume over time, from a set of independent variables expected to affect it. The equation shows what a change in each variable means for the KPI. Meta gives this high-level flow:

1. External factors and marketing inputs.
2. Impression and business metrics.
3. MMM statistical regression analysis.
4. Data outputs.

Meta notes that exact MMM methods vary by locale, vertical and company.

What are the steps in an MMM study?

1. Define the business questions and scope, so the design matches what you need answered.
2. Collect time-series data for sales or another KPI, media, non-media marketing such as promotions, and macroeconomic factors.
3. Review the cleaned data for accuracy before modeling, which may reveal gaps that need more collection.
4. Run the models, expecting to re-run and fine-tune them several times.
5. Analyze the outputs and make recommendations tied to the original questions.

Meta stresses that inaccurate or poor-quality input data produces inaccurate and poor-quality outputs, and calls data collection one of the most important and time-consuming steps.

How can you build one?

Route	What Meta describes
Third-party MMM vendor	Specialists who usually run the whole engagement, from discovery to recommendations; most offer software for simulation, optimization and forecasting
Semi-automated tools or SaaS	Point-and-click modeling needing less specialist statistical knowledge, with automated modeling that can run on an ongoing basis
In-house build	Your own team runs and manages everything internally

What is Robyn?

Robyn is an MMM project covered in Meta’s help centre. Meta says it provides end-to-end modeling scripts that can be used whole or in part in other MMM projects. It aims to increase automation and reduce human bias, but Meta is clear that it doesn’t run without intervention: analysts still tune parameters, try different modeling techniques and interpret the results.

What questions can MMM answer?

Meta lists questions such as:

• How many online and offline sales did each media channel drive?
• What was the ROI of each marketing channel?
• How should you allocate budget by channel to maximize your KPIs?
• What is the optimal spend level for each major channel?
• How should you reallocate if you must cut the budget by a set percentage?
• How much incremental revenue comes from non-marketing activities?

FAQ

Is marketing mix modeling the same as attribution in Ads Manager?

No. Ads Manager attribution credits conversions to ads within its attribution model. Meta describes MMM as a holistic regression model across marketing and non-marketing activities over time, used to estimate each one’s incremental impact and ROI and to plan budget across channels, products and regions.

Does Robyn build my model automatically?

Not on its own. Meta says Robyn aims to increase automation and minimize human bias, but that analysts still need to fine-tune parameters, run different modeling techniques and interpret results to produce useful recommendations. Meta also suggests collecting your business questions first to check MMM fits.

What data does an MMM need?

Meta lists time-series data for sales or another KPI, media activity, non-media marketing such as promotions, and macroeconomic factors. If you want budget advice by channel, Meta says you need granular data at the executional or creative level as a model input.