Why you’re probably getting attribution wrong (and what you should be doing instead)
When it comes to digital marketing as a whole (and display advertising specifically), accurate attribution is the holy grail of data.
When you understand exactly where sales come from and why, you can plan and replicate successful campaigns with confidence, and stop wasting money on marketing activities that don’t deliver.
But accurate attribution is incredibly difficult to pull off – and few companies or even agencies get it right. So we thought it helpful to look at some of the ways brands are approaching it, and the strengths and weaknesses of each.
The no attribution approach
The simplest way to approach attribution is not to have any at all, and instead rely on different media owners and channels to tell you how well they performed.
For example, if you had a paid search provider, four display providers, an affiliate provider and SEO measurement, each channel/media provider would report how they performed, how many sales or other actions they drove and ROI.
Sounds easy? The problem with this is that different channels and media owners will all claim the same sale. The attribution used by each channel/media owner in this case is likely to be very basic, so they won’t really have any idea of how many sales genuinely came from them. In this instance they’ll naturally want to try to claim as many claims as they can.
As a result, what you’ll normally find with the no attribution approach is that the total of all the ‘sales’ from all your different media owners will be vastly greater than all the sales that occurred on the site.
If you do decide to use the no attribution model, you need to do it directionally, as it won’t show a true CPA or ROI. Use directionally however, it can in most cases show which media owners perform better than others.
The strengths of no attribution
- It’s easy to setup
- It can be used to assess directionally which media owners are doing better
The weaknesses of no attribution
- It tends to dramatically over count sales
- It’s unable to reliably show how much value each channel or media owner produced
- It leads to multiple tags on your site, which can put you at risk of data leakage
The last click or last impression attribution approaches
The two most well-known and used attribution methods are last click and last impression. This basically means that a sale is attributed to the last click or impression before the user converted.
Unlike the no attribution approach, both these models only count each sale once. However, that is where the good news ends. Why? Because last click and last impression attribution both tend to over assign value to the bottom of the funnel (typically either retargeting or brand search), which means that the bottom of the funnel is overworked from a strategic perspective.
Last click attribution also does not take into account the visual impact of a display ad, relying solely on clicks, which results in display being massively undervalued. As a result, display prospecting, generic search, content affiliates are all underused tactics across the industry.
The strengths of last click and last impression attribution
- They’re widely available methodology
- They ensure each real sale is not double counted
- They’re inexpensive to use
The weaknesses of last click and last impression attribution
- They don’t attribute to where the true value is produced
- They can massively over value channels and tactics at the bottom of the funnel
- They massively undervalue tactics further up the funnel including display prospecting, generic search and content affiliates
- Last click attribution particularly undervalues display, as it does not take into account the value of viewing an ad, only clicking upon it
The fractional attribution (click and impression) approach
The fractional approach to attribution assigns the value of a sale to different parts of the user journey.
Some fractional attribution models include first click, impression, linear attribution (where each step in the chain counts evenly), and time decay (where it assigns more value to an engagement the nearer to the time of sale you go).
The advantage of these models is that they can remove the last touch tendency to push the bottom of the funnel, making them a big step forward for attribution and an advertiser’s strategy.
However the challenge with this approach is that, whichever model you choose, you cannot be sure that it will be assigning the value where it actually belongs. For example, if you give all parts of the chain the same weight then you’re not taking into account that some parts of the journey are more valuable than others.
This model can be very effective when combined with a strong strategic approach to attribution and media buying.
The strengths of fractional attribution
- It ensures each real sale is not double counted
- It allows all parts of the funnel to have a chance to be attributed
- It’s relatively inexpensive
The weaknesses of fractional attribution
- It doesn’t get you to the “true value of each media channel or touchpoint
- There’s no guarantee the new fractional model is actually better than last click or impression
The baseline attribution approach
A baseline of sales is the number of sales that would have occurred had no marketing activity taken place (or no digital activity). The baseline approach to attribution measures your baseline conversion rate, then shows how each touchpoint or channel influences your chance of converting over and above that base.
Baseline attribution systems are based on the likes of probability, machine learning or game theory to predict your chance of converting, and calculating how each piece of media influences that chance.
Different vendors will choose different mathematical techniques for this, but all essentially rely on analysing all users exposed to marketing activity for an advertiser, and not just those that convert.
These systems promise to show the true value of each piece of media, though it is hard to verify that this is actually the case. Instead, judging platforms should be done on increasing strategic understanding, and through applying baseline attribution techniques to show that the advertiser is able to run more effective activity.
Baseline attribution models are the best available for most forms of digital marketing, however they require a huge amount of number crunching, making them too expensive for some advertisers.
The strengths of baseline attribution
- It promises to deliver the true value of each piece of media
- It’s the best form of digital attribution available for larger advertisers
- It potentially allows for more effective and strategic media campaigns
The weaknesses of baseline attribution
- It’s too expensive for many advertisers without big budgets
- It’s hard to verify the accuracy of any systems numbers
The strategic approach to attribution
The cost of baseline attribution can be prohibitive for some advertisers. Instead, the best approach for these advertisers is to combine a strategic approach to whichever attribution system can be justified. Ideally that would be fractional attribution, but it can also be effective on last click and last impression models if it’s applied correctly.
This means that rules need to be created to override the weaknesses of the chosen attribution model. Here are some examples of rules that are used:
1) The split attribution model – this model attributes each sale twice: once to whoever was the touch before sale, and once to whichever media was the last touch before the first visit. This allows both the bottom of the funnel and the middle of the funnel to be rewarded for response to sales. You can take this approach further by only giving half the sale to each, and splitting even further through multiple touchpoints. This model can be done with quite simple technology and is easy to understand.
2) Only allowing for attribution viewable impressions – this prevents the un-viewed cookie bomb techniques where cheap and un-viewed impressions are served at volume to take credit of sales. This is very effective but does need specialist technology and data processing.
3) Choosing which channels should be attributed against – we don’t believe that brand search, for example, should be treated as a media channel because brand search is really just used for lazy but effective navigation by most users. So the last click before brand search would get the sale.
4) Assigning different targets to different parts of the funnel – for example, prospecting should be given a much higher CPA goal than retargeting. Equally, generic search would have an easier goal than brand search. This technique can be taken further with multi touch attribution, ROI and lifetime customer value metrics.
These four examples are just a snapshot. Strategic use of attribution can be very effective for media at a limited or even no extra cost to run. The challenge is that it requires expertise to setup, run and understand. This means not all aspects of the business may buy into the approach.
Equally, it’s hard to prove that any given approach is optimal and is more down to trust that this approach is best for the business. You can overcome these concerns with continual testing, auditing and value cases, but these again add complexity to the narrative.
The strengths of strategic attribution
- It has a low to zero marginal cost on technology
- It allows for highly effective media buying
The weaknesses of strategic attribution
- It requires expertise to set up
- It’s difficult to communicate why a given model was shown to those without a deep knowledge of digital media
- Media may look to be underperforming to certain parts of the business only using the simple attribution systems and without the strategic background knowledge
- It’s hard to implicitly prove that any method is the right one, so it requires trust in the strategic direction
The econometric approach to attribution
The econometric approach is a very different form of attribution that evaluate all media channels, not just digital. It’s the established method of measuring all media together, from TV, out of home (OOH) and print, through to digital display, search and everything in between.
The econometric approach involves comparing media spends in each channel to a baseline amount of conversion. Econometric maths techniques are used to show how a given increase or decrease in spend leads to an increase or decrease in sales. As such ROI calculations and optimum spends for each channel can be calculated.
We could go into great length and depth about econometric techniques, but there’s not the space here! However, we do want to stress that econometric modelling is the best form of measurement for offline media and will play a big part in how the best marketing plans, cross channel, are evaluated.
The strengths of econometric attribution
- It has a low to zero marginal cost on technology
- It allows for highly effective media buying
The weaknesses of econometric attribution
- It requires expertise to setup
- It’s difficult to communicate why a given model was shown to those without a deep knowledge of digital media
- Media may look to be underperforming to certain parts of the business only using the simple attribution systems and without the strategic background knowledge
- It’s hard to implicitly prove that any method is the right one, so it requires trust in the strategic direction
What attribution approach are you using?
So what attribution approach have you been using for your display advertising? And, after reading this article are you confident it’s the right choice for you?
If you’d like advice on your display advertising as a whole, or attribution approach in particular we’d be very happy to help. Just email us on jweeks@crimtan.com or call us on 020 3262 0415.