This article is a reproduction of the types of attribution featured in the Econsultancy whitepaper “State of Marketing Attribution in Asia Pacific”.
You can download the full study here.
As the diversity of channels, devices and touchpoints has increased over the last few years, the challenge of attribution has become more complex. Along with this diversity has come the era of big data, allowing a level of customer journey analytics not previously seen. The application of big data into attribution models has developed the range of methods used by companies.
Despite this, Figure 10 shows that first touch or click, which many would class as a basic form of attribution, remains the most commonly-used model beyond last-click according to both companies and agency respondents (47% and 50% respectively). Encouragingly, second to this is custom modelling, used by 39% of companies, and a higher 49% of agency clients.
Custom attribution modelling uses one or more standard models as the starting point, and then layers in other factors unique to a business to produce a custom model. Though complex to set up and monitor, the method produces the most relevant model for every business, ultimately increasing accuracy.
The 26% higher use of custom modelling by the clients of agencies indicates the expertise of the latter when it comes to attribution technologies and methodologies. Concerns are regularly expressed about media agencies owning attribution, in that there is a danger of them ‘marking their own homework’. However, agencies certainly have the experience and internal skills to make the most of their clients’ big data, using more complex models than perhaps companies are able to in-house.
Even allocation (linear) attribution, where each touchpoint along a journey to purchase is given equal value, is the least-commonly used method, with just over a fifth (21%) of companies selecting it.
The increased accuracy of custom modelling is reflected in Figure 11, which clearly shows the augmented effectiveness of the method, with 41% of companies rating their custom attribution model as ‘very effective’, and a further 41% rating it as ‘somewhat effective’. In comparison, even allocation (linear) was rated as ‘very effective’ by none of those surveyed, and as ‘somewhat ineffective’ by more than a quarter (27%).
Agencies rate even allocation as slightly more effective than the client-side, with 71% rating the method as ‘somewhat’ or ‘very’ effective. However, custom modelling is again viewed as the most effective method, with almost half (49%) of agencies saying their clients rate it as ‘very effective’.
Survey respondents were asked in an open question if any attribution type or approach has proved particularly effective, producing a mix of opinions (box overleaf). Some are sticking with simple, click-based approaches, either because they are just starting out using attribution models, or because they don’t have the internal resources to progress to a more complex model. Others were evidently further ahead with their attribution modelling, using custom modelling to optimise marketing budgets across both publishers and channels.
The complexity of attribution modelling is reflected in Figure 13, which shows that 64% of respondents agree that a perfect attribution model is impossible to achieve. The attribution trends briefing based on this year’s Digital Cream event1 discussed the reputation that attribution has gained as being ‘divorced from reality’, stating:
“Sometimes there is scepticism about the amount of arbitrary rules in place, which means that errors can be introduced into the modelling. It’s often seen as just one version of the truth within companies; the definitive view of the world. However, it’s a mistake for companies to try to build the perfect model as ‘perfect is the enemy of good’.”
The vast majority of experts believe that attribution is not a perfect science and it can’t solve every question a marketer has about their campaign mix. However, they do help marketers to evaluate the influence of different channels on each other, and the comparative impact of each on conversion rates.