Multimedia Mix Modeling: Why Smart Advertisers Are Using it to Overcome Tracking Challenges
With the gradual deprecation of third-party cookies, online advertising measurement has become more difficult (offline advertising measurement always has been – without tracking, how can you measure the impact of traditional radio or TV ads?).
In this context, here at Blackbird PPC, we have been finding a lot of success using offline measurement methods for online initiatives. More specifically, I am talking about Multimedia Mix Modeling, aka MMM. In my experience, MMM works well as long as you have enough historical data to look at.
Let’s examine why MMM is gaining momentum on the marketing landscape.
Click-based reporting is less and less reliable!
Most advertisers have been observing this phenomenon over the past few years. The share of unattributed sales has been increasing as third-party cookies are going away, making it harder and harder to measure channel-level performance. Unfortunately for marketers, multi-touch attribution is fundamentally flawed.
MMM can help you understand the true impact of your investments
While geo lift testing allows to measure the impact of specific initiatives through online controlled experiments, MMM allows to account for all initiatives at once and helps marketers understand the main predictors across paid media, organic initiatives, as well as holidays and seasonal trends.
Lately, we’ve been using Meta’s Robyn, a semi-automated MMM open source package. What is great about it is that, because the inputs of the model don’t require channel-level revenue, holidays and seasonal trends are accounted for, and it doesn’t matter if your click-based tracking is broken or unreliable! here are the required inputs to run Robyn:
Channel-level spend
Aggregated revenue, as opposed to channel-level revenue
Optional: any additional organic or context variables such as newsletters, promotions, events, weather, unemployment rate, COVID, etc…
With Robyn, aggregated revenue can be attributed to every single channel irrespective of any click-based tracking challenges, and MMM results will only get better as the model is fine-tuned with the addition of organic and context variables. The below view shows an example of how revenue may be attributed across channels:
You can assess MMM’s accuracy through cross-validation
Typically, cross-validation allows us to compare counterfactual predictions vs. actual performance, such as in the below graph. The comparison tells us how strong model accuracy is – hopefully, predicted and actual performance are aligned! If they aren’t, that means you definitely need to modify your inputs, whether it is more organic or context variables or potentially more historical spend data.
At Blackbird PPC, we’re well versed in using MMM to tackle marketers’ measurement challenges, so feel free to reach out if you’re interested in learning more.
