The key to Successful Media Mix Modelling
Media mix modelling (MMM) is an established statistical technique in the marketing world to calculate the impact of media campaigns. Aside from helping understand the effect of general media campaigns, the technique can be used to measure the effects of pricing, promotions and events such as product launches. Getting this holistic view is one of the main advantages of MMM. More recently, the multi-touch attribution (MTA) model has been favoured over the media mix model. The MTA model provides the more detailed, necessary insights into customer journeys within specific channels and campaigns. This is a more granular approach than the general, channel-level MMM. But, MMM is far from outdated. The MTA model falls short for offline channels such as television or radio advertising, as most of this data is not available at an individual level. MMM plays an important role in determining the impact of these channels, and herein lies its continued relevance.
This blog will discuss when to use MMM, how to set up a good media mix model and how to combine the results with those of the MTA model.
When do I use media mix modelling?
At a time where conversion attribution is finding its way into many marketing departments, you might find yourself asking why you still need the channel-level MMM. For many channels, MTA is indeed a great technique to study the marketing effects on an detailed, path-based, level. However, offline channels cannot be included in these paths. There is no way to tell whether someone saw your television advertisement or heard your radio campaign. This offline data is aggregated, cannot be tracked on an individual basis and can therefore not be included in the MTA model. MMM can fill this gap. It can be used to incorporate media investments that can only be measured on an aggregated basis, thus allowing the comparison of all media investments and presenting you with a holistic view of all your marketing activities. The comparison of the upper and lower images below illustrates that a television commercial can induce new touchpoints in existing customer paths, start new paths and increase a path’s chance of conversion. The images below also show not all customer paths are affected by the television commercial, as not every individual has necessarily seen it. MMM can be used to quantify the aggregate effect of the commercial.
How do I set up a good media mix model?
- Make sure you have good-quality data available. MMM is also most effective when using a large dataset. You need a minimum of two or three years’ worth of data for solid results.
- Select which KPI you want to measure. This could include website visits, attributed conversions or attributed revenue as measured by MTA, for example.
- You also need to set up your independent variables, consisting of the variables that affect the selected KPI. Imagine your selected KPI is the number of website visits per day. Have you ever thought about how the weather affects the search behaviour of your customers? Or how a salary payment might trigger a sale? These contextual factors that can influence your KPI need to be included in the media mix model. Additionally, a good media mix model needs to capture changes in the market. Events such as sale periods, competitor advertising, and website failures also need to be modelled. Including your offline media (television, radio and out of home) completes the model.
- For this last set of data it is particularly important to use the right modelling techniques. Taking into account the advertising adstock model is key to a good media mix model. A consumer will partially remember a commercial they see today tomorrow and the day after that. Over time the knowledge of the commercial fades completely. If a company broadcasts many commercials during a given flight, the knowledge stocks up and stays in consumers’ minds for a longer period of time.
How can I combine the outcomes in the MTA model?
Now that you know the effects radio and television campaigns have, they can be included in your existing solution. On one hand, the outcomes of your MTA model show you how one conversion should be attributed over various online conversion paths. On the other hand, the outcomes of your media mix model show the additional conversions, revenue or website visits your aggregated offline media channels caused. The challenge you now face is to re-attribute some conversions that you initially attributed using MTA in order to combine the information given by the two models. The conversions are reattributed proportionally to the impact each channel has. Consider the following simple example of conversions related to a television campaign:
Imagine you first ran the MTA model, which attributed 50 conversions to the paid search channel. The output of your media mix model subsequently shows that 10 of these paid search conversions are a result of your television campaign. In this case, one fifth of the paid search conversions can be re-attributed to the television channel. This can be done for all channels, the sum of these re-attributed conversions yielding the total effect of your television campaign.
The figure above illustrates this process. The conversions have been attributed using MTA and subsequently form the KPI for the media mix model. The yellow bars show the gross rating points (GRPs) of television. The media mix model attributes a certain amount of conversions to television, as shown in dark blue. These conversions were previously all attributed to the online paid search channel by the MTA model, but are now partly re-attributed to television advertising by the media mix model. Following the example above, one fifth of the €25, €5, goes to television:
The biggest added value of MMM is that it enables you to measure the impact of those channels MTA cannot include. Subsequently combining the results of MMM with the attributed conversions of your MTA model gives you a complete overview of all your media investments. Once the results of both models have been combined, it is easy and acceptable to compare the effectiveness of media spend across channels. Aside from this holistic marketing view, MMM can give you insights into the impact of events such as sale periods and product launches. A good media mix model is therefore crucial to measuring the effectiveness of campaigns and allows for an accurate comparison across all channels.