There is a current school of thought within advertising that marketers over-value the contribution of online marketing in driving business outcomes, while the benefit of offline marketing and brand-building activities is not clearly visible to marketers and as such is being incorrectly attributed to short-term response-driving techniques.
There are many reasons why this is happening. The most obvious is that the average tenure of a CMO is now just four years, meaning the key decision maker often isn’t around to see the benefits of building their brand over the long term.
But I believe there is a bigger and more fundamental issue around measurement, specifically the number of businesses who rely on digital attribution models to measure value and ultimately inform future media planning.
The fallacy of data-driven attribution
A good snapshot of how marketing attribution technology is being used comes from AdRoll’s State of Marketing Attribution report. It shows that last click is still the dominant methodology being used by both brands and their agencies, with 48% of companies stating this was the primary attribution method. For agencies, the figure is an even higher 58%.
It has already been widely acknowledged how last click as a metric is flawed and overly simplistic.
These algorithmic and custom attribution models are often labelled as being ‘data driven’. The reality is that these models are data-driven, but only driven by specific types of digital data which are inherently biased towards digital media.
When you scratch the surface it becomes clear that, like last click, they too are overly simplistic and suffer from many of the same flaws in logic. Yet marketers have become hooked on the quick fix that these models provide.
Whilst the exact attribution weighting between channels is set by the attribution algorithm, sales credit is only ever given to digital media touchpoints that customers see or click on prior to a sale happening – as long as those interactions took place within the cookie window. Yet the real kicker is that all digital sales are credited to some form of digital interaction, regardless of whether media played an active role in converting the user.
To be deemed successful, media plans need to deliver against these new algorithmic measures of success. Media planning and optimisation becomes a form of cookie chasing. With the prevalence of data-led media buys, it is easier and lower cost to deliver ads to users that are already likely to buy than it is to create new buying behaviour.
Speaking to high propensity users is over-valued in these models, whereas speaking to low propensity users is undervalued. After all, you can count more total sales in a siloed channel when speaking with the high propensity audience, but this ignores that many of these users are part of the ‘baseline’ and are therefore not incremental to the business.
Propensity as an enabler of marketing attribution
In my view, the biggest flaw in digital attribution logic is that the models assume that there is a causal relationship between media exposure and sales, yet we know from econometric modelling that this is not the case.
For established brands, the largest driver of sales is invariably ‘baseline’ – those being any sale that was not directly caused by media – but you will almost never see baseline quoted in any digital attribution reports. Instead, sales are credited to a variety of paid and organic online media channels.
To truly understand media effectiveness, it is important to recognise that there are many different customers at different stages of the buying cycle. To know whether media has worked, it is vital to know what was likely to have happened had the user not been exposed to advertising – we need to draw a distinction between sales that were ‘tracked’ and those that were ‘caused’ by media.
In recent years, the most successful campaigns I have worked on have all used a basic version of this logic, with success measured on incremental sales rather than total tracked sales. With the prevalence of digital data, it is possible to gain a basic understanding of a consumer’s propensity to purchase based on their digital footprint. A simple approach might map propensity as follows:
High propensity users will exhibit strong buyer signals such as brand or product search, or be part of a retargeting pool – having previously visited the advertiser’s website.
A medium propensity user will demonstrate that they are ‘in market’ for a given product or service as determined by searching for generic keywords, competitor brands or from doing online research around the brand’s category or sector.
A low propensity user is everyone else, i.e. those that have not demonstrated any form of positive brand or category purchase intent.
Where a brand has customers with repeat purchase behaviours, this data can be used to further predict purchase behaviour, therefore enriching their view of purchase propensity.
In taking this approach it encourages marketers to think in terms of sales uplifts rather than the more basic measure of tracking sales. When advertising has a causal relationship with sales this can be recognised as results that are mitigated against the existing sales baseline. The result is that the business starts to think in terms of generating new and incremental consumer behaviour, which unlocks business growth.
Applications with AI & machine learning
Of course, this basic propensity approach can be further enhanced and improved. There has been a great debate in our industry about how artificial intelligence and machine learning will impact the future. The narrative has focused on the automation of media buying, but I believe there is an even larger opportunity to automate the quantification of propensity data at scale.
When combining consumer propensity data with a rich, data-driven attribution model there will be a measure of success that recognises a causal relationship between media spend and sales. It is only at this point that brands and their agencies should feel comfortable in relying on digital attribution data to determine media effectiveness.