Marketing attribution and its relevance have become a hot topic of conversation, with discussions comparing the outputs of attribution with econometrics or marketing mix modelling (MMM), and raising concerns about the ‘variance’ and ‘correct’ approaches to marketing measurement.
It’s time to address why this comparison exercise is both naive and irrelevant. Measurement techniques are not built to compete, they are designed to complement. There is no one-size-fits-all solution. How brand owners harness the diverse capability and complementary power of a holistic marketing effectiveness suite will vary depending on the individual needs of their business.
Rule-based vs data-driven attribution models
Confusion reigns with the term ‘attribution’, which describes a broad range of measuring approaches. A curse that attribution modelling endures is that the term was overused when Google popularised its last-click method. This gives all credit for a sale to the last thing a customer does before purchasing.
This same term then evolved to describe rule-based models, which similarly divide the reward for generating the sale among the events immediately preceding It, based on a subjective decision. While these models provide helpful information to marketers, they suffer from a lack of objectivity. They also promote short-term and siloed thinking.
The ambition of attribution modelling is that each event in a customer’s journey to purchase is rewarded with the value it drives. This provides powerful insight into the role and impact of marketing. Marketers want this view but need an unbiased, holistic approach that includes digital channels.
Data-driven attribution modelling has answered this demand. Data-driven models provide an unbiased view by using mathematical techniques to determine each marketing action’s incremental impact. These models, also known as custom attribution models, can be tailored to a brand’s specific objectives, segments and insights. They can also provide a broader perspective by covering a longer purchase journey. They are typically more sophisticated and complex to build, but provide much more accurate insights into marketing performance.
Unlocking the full potential of attribution
When considering data-driven attribution modelling, it’s important to consider incorporating the following capabilities to maximise the power, accuracy and usefulness of the insights:
- A baseline measurement is needed to ensure incremental marketing impact, taking account of all the external factors affecting sales.
- Accounting for all exposure to marketing by measuring impressions as well as clicks and employing impression modelling techniques where the actual data Is not readily available.
- Connecting all marketing activities including non-digital touchpoints such as direct mail and email, to get a complete view of the customer journey.
- Including the direct impact of above-the-line media such as TV and radio on digital activity in attribution models, using a sophisticated spot-matching approach.
Marketers should be aware of the benefits of a data-driven customised attribution model that incorporates these features, compared to more simplistic attribution techniques, when making comparisons between the approaches.
The ‘golden trinity’
Comparing attribution modelling with econometrics is like comparing apples and oranges. Attribution’s role is to understand individual customer behaviour and journeys, while econometrics is about understanding the full impact of market and macro factors alongside an aggregated channel view of sales drivers.
When employed appropriately, data-driven attribution models have a beneficial and important role in any measurement framework but are not comparable to econometrics. They don’t answer the same questions or use the same techniques. Econometrics is strategic, long-term and macro; whereas attribution is tactical, mid-term and micro.
The best way to properly measure effectiveness is for attribution, experimentation and econometrics to evaluate performance together.
In previous articles on Marketing Week, we have emphasised an urgent need for marketers to adopt a more agile and holistic approach to measurement, employing multiple metrics from different sources to form a sensible and coherent story around marketing’s impact throughout the funnel, and the combined contribution of the marketing mix. This requires a more sophisticated approach, overcoming the siloed mentality that has been nurtured by more simplistic tactics, using multiple data sources and various measurement techniques to gain a comprehensive view on effectiveness, and overlaying human context to make sense of the results.
By classing one approach as ‘wrong’ and one as ‘right’, the benefits of a holistic measurement system are overlooked. Instead, the preferred approach should be to employ the solution that works best for the requirement while being cognisant of the limitations of this choice.
Google’s recent paper supports the view that the best way to properly measure effectiveness is for attribution, experimentation and econometrics to evaluate performance together. Experimentation can validate where modelling outcomes differ, and the models are able to incorporate the results of uplift tests in their assumptions. This allows the appropriate measurement techniques to be used based on the types of decisions that you need to make. Effectiveness expert Les Binet recently described this as “triangulating ROI”, while others refer to it as ‘unified marketing measurement’ or the ‘golden trinity’ approach.
The combination of these three techniques creates a durable system, which can adapt to any inconsistencies through each learning from the others. This allows the golden trio to be more focused.
Econometrics provides a longer-term strategic outlook of sales drivers, and can forecast budgets and outcomes, but is constrained in its capabilities to look further – for example at a campaign, creative and keyword level. Attribution can then be employed for monitoring customer touchpoints and the short-term, granular media channel efficiencies of these budgets. Experimentation builds on the incremental views of each approach to support in-flight tactical decision making and overcome any discrepancies.
Studio Retail has already seen the benefits of using this holistic measurement approach on performance, by employing both econometrics and attribution modelling alongside experimenting with spend through Google.
Innovating for the future
The need for agility in marketing measurement is growing, as traditional methods prove slow and costly. Data gaps – caused by the demise of cookies and privacy laws impacting how data can be collected, in addition to walled gardens – will also limit measurement capabilities, potentially forcing marketers back into less inclusive last-click approaches. Channel reporting that sits alone outside of any framework will become increasingly ineffective.
To address this, marketers are seeking a balance between current methodologies and faster, lower-cost alternatives that still retain the outputs needed. Open-source model projects such as Robyn by Meta and LightweightMMM by Google aim to automate measurement and build models quickly, but still require a modelling skillset and development over the long term to become as feature-rich as marketers require. Those choosing to use these approaches should not underestimate the validation time required over the long term as new capabilities are built, bugs are fixed and updated code bases are released. However, it’s a step in the right direction to bridge the data gap.
Comparing different modelling approaches is futile
Econometrics and attribution methodologies have very different roles to fill and they supplement rather than replace each other. A holistic measurement framework should be driven by channel mix and objectives, bringing together a variety of outputs to reach an informed view.
As data becomes increasingly restricted, measurement model approaches will be integrated to overcome this gap. Modelling will always come with some ambiguity, but in today’s data-limited world, it’s now more important than ever to get comfortable with evolving measurement solutions and use human context rather than the data alone to explain their outputs.
Hanna Wade is director of effectiveness at Jaywing.
Need a different kind of agency? Let’s talk.