GLOBAL: Achieving a total view of attribution is an iterative process which requires a combination of econometric modelling and digital attribution, according to an industry figure.
In a WARC Best Practice paper, How to develop an attribution model for your brand, Patrik Sahlin, a principal consultant at Ebiquity, offers a practical introduction to these two main branches of attribution, discussing the steps that are taken to build an attribution model and how to choose between different methodologies.
“The fundamental difference between econometric modelling and digital attribution is the change of view,” he explains.
Thus, econometric modelling works with aggregate data and measures marketing effectiveness from the brand’s point of view, while digital attribution uses transaction data and takes a consumer-centric view.
Econometric modelling is used to model the marketing mix and help marketers to identify drivers of key performance indicators and to quantify their effects.
Sahlin outlines how “The incremental effects of marketing activities such as media campaigns, price, distribution, and promotion” can be measured “and factors such as seasonality, weather impact, or base sales can be separated”.
Digital attribution, meanwhile, uses transaction data and can measure the effectiveness of digital media channels with greater detail than econometric modelling.
Both techniques have strengths and weaknesses that need to be considered, including data collection and validation.
Econometric models require data from multiple sources, some of which will come from outside the brand’s organisation and which it is the analyst’s responsibility to validate.
The industry standard for econometric modelling is multiple linear regression, Sahlin adds, and marketers should understand what assumptions are made by external vendors and what the limitations of the methodology are.
For digital attribution, browser cookies have been central to following a consumer from first interaction with a brand to conversion, but these cannot account for multiple users of a single device or a single user moving across multiple devices.
Walled gardens, such as Facebook, add another complication as they don’t allow for tracking across their domains.
In deciding where to start, two main factors must be taken into consideration, Sahlin suggests: are sales online or offline, and what share of the media budget is digital (digital attribution should only be an option if digital media has a significant proportion of the media budget)?
Ultimately, econometric modelling helps to evaluate and optimise the top-level media budget, including offline and online media channels, while digital attribution helps to optimise and improve digital media channels.
Sourced from WARC