Unified Attribution Modelling
Marketing managers want to know how to invest their budget. Ideally, they want to invest in channels with the highest incrementality; the channel that causes the most uplift in conversions and revenue.
How do you determine how much uplift was caused by each channel?
Example: a customer gets an e-mail, then a Facebook impression and finally looks up the article via paid search and converts. How much value from this conversion can be attributed to each of these channels?
This is the area of attribution modelling. There are two established attribution models: multi-touch attribution (MTA) and media mix modelling (MMM). In the market we often encounter a preoccupation with which model to use and whether MTA is better than MMM or vice versa. This is not a relevant question, however. The business question – how much uplift was caused by each channel – is leading. The model you then use depends on which channels you are measuring, as depicted in the flowchart below.
For online channels, most touchpoints can be measured at the user level and you can track which touchpoints were shown to your customers, in which order. You can accurately model the value of each touchpoint using MTA. Depending on your tracking solution, you might only have aggregate data for some online channels like Facebook or display. If you only have access to aggregate data, you should use MMM. As soon as you have both user-level and aggregate data, a unified model combining MTA and MMM should be used to leverage the strengths of both models.
In general, MMM can be used to measure the impact of all your channels, but not at a granular user level. MTA is much more granular, but is limited in its capacity to measure traditional media and online impression channels. The unified model is both granular (where it can be) and holistic.
Consider the following example: a customer sees an ad on Facebook and then decides to take a look via paid search. Since you don’t have a tracking solution, Facebook impressions are available on an aggregate level; you are not sure which user saw which Facebook ad and you can’t place these impressions in individual customer journeys. In this case, you have a mix of user-level and aggregate data. If you would use MTA, the lack of user level data would make it seem as if the customer went straight to paid search, giving paid search all the credit for the conversion. Paid search is overvalued and the impact of Facebook is ignored. Measuring the impact of Facebook on this conversion thus requires MMM. Only by combining MTA and MMM in the unified model will you eventually get a complete view of the value of all channels.
How does it work?
Imagine a customer with the following converting path:
First, MTA divides the credit over all channels in the path that can be measured at the user level. MMM then decides how much credit the Facebook impression deserves, but also indicates which channel is overvalued by MTA, SEA in this example. This is then corrected by re-attributing part of the credit to Facebook, as shown in the table below.
|MTA + MMM:
The attribution model you choose thus depends on whether your data is available on a user or on an aggregate level. Subsequently combining the MMM results with MTA gives you a holistic overview of all your media investments. This allows for a comparison of the effectiveness of media spend across all media channels.