Media Mix Modeling, How We Crack Your Business’s Unique Code
The world of media buying is constantly evolving. There is a plethora of channels across both digital and traditional media, in which to invest marketing dollars. The rise of retail media, new social media networks, subscription services like Netflix now allowing ads, continued adoption of connected television (CTV)… It's hard to keep up! On top of that, digital attribution is becoming less certain with the death of third-party cookies — the bit of code that allows us to track if website users came from ads. How do you know what channels are right for your business goals? How will small-to-medium sized marketers attain true attribution without access to hyper-expensive products like LiveRamp or relying solely on first party data? At Lewis Media Partners, we are integrating media mix modeling (MMM) as an integral part of our attribution solution.
MMM is a statistical attribution model that provides a holistic picture of what factors affect business outcomes without compromising privacy. Rather than requiring personally identifiable information (PII), it ingests actual business results (revenue, acquisition), ad spend, and external factors such as macroeconomic indicators or flu cases. Incorporated data needs to be time series data, a sequence of data that is arranged over time intervals like week or day. A familiar time series to all of us is the DOW Jones Industrial Average. Those graphs you see of changes in the DOW over time? That’s time series data.
By analyzing time series and understanding their correlations with each other along with the actual business results, MMM creates an attribution model. This attribution model can assign a true return on investment (ROI) or cost per acquisition (CPA) to each factor in the model -–- whether paid media in the digital or traditional space, organic media like the reach of your Facebook posts, or external factors like consumer price index (CPI – a common index for US inflation). This will allow us to understand not only how each media channel is performing, but also how external factors affect your business.
With this attribution model, we can parse out optimized media spends for each channel. We know media channels do not scale infinitely, but what is the exact point of diminishing returns? How can we squeeze each channel for all that it’s worth? MMM gives us diminishing return curves, and it optimizes budgets for each channel to provide a media mix that will maximize efficiency. I want to stress the word “efficiency”. In our current period of high inflation, being efficient is more important than ever. Your ad dollar gets you less and less every day, - and marketing budgets are not keeping up. At Lewis Media partners, we work with what you have to drive results.
MMM provides us the ability to create budget scenarios. What happens if we use the optimal media mix? How should media investments change if we find a particular macroeconomic factor that is wreaking havoc on your business? One of our clients decided to shift dollars into brand awareness tactics vs. performance tactics based on MMM results. While that client’s industry is hurting, they are faring much better than competitors and continue to gain market share. MMM empowers us to shift strategies, tactics, and budgets to align with cyclical patterns and advertise in the right place, with the right message, at the right time.
Why are we employing MMM in 2023? Our most seasoned marketers have heard of MMM — Kraft implemented MMM to optimize their Jell-O campaigns in the 60s and 70s! However, what marketers have not had in the last 50 years is cheap computational power and artificial intelligence (AI). With our current tech stack, we can produce hundreds of thousands of models an hour, lump together similar models using AI, and then use human intelligence to determine the best model.
So how does MMM work? In a nutshell, we take all our ad spend and external factor time series data (independent variables) and create an equation for the actual business result (dependent variable) using regression; kind of like the equation of a line: y = mx+b. We can then take that equation and measure each factor’s impact on the business.
We utilize ridge regression in our MMM analyses. Unlike linear or multiple linear regression, ridge regression assigns a penalty term to each factor in our model. That means any of our independent variables that are correlated will have a penalty assigned to them. We also utilize statistical methods for determining adstock as a factor in our modeling. Adstock is the idea that advertising can have a carry-over effect on consumers. Are there any ads that stick in your head after you’ve watched them on Hulu or cable television? That’s adstock.
MMM is an essential tool for understanding true attribution, optimizing media budgets, and driving results in a post-cookie world. By analyzing the performance of different media channels and external factors over time, we can identify underperforming channels, make data-driven decisions that improve marketing’s effectiveness, and plan your media budgets to be more efficient and effective. Please feel free to reach out to me at email@example.com to talk about how we can help you and your business prepare a true attribution solution using media mix modeling.