For a long time, attribution modelling was the number one choice to analyze the impact of each channel in your marketing mix. And this feels true even today - You’ll see it in most of your advertising and analytical platforms.
But there was always the question - Is it really the best way to analyze the advertising performance?
In order to understand where this question is coming from, we have to start with understanding the attribution modelling itself.
The customer journey, especially today, will most likely include multiple channels and there’s no denying that one of the key factors to a successful digital marketing campaign is understanding what works and what doesn’t.
How is it possible to find it out?
Attribution modelling offers an easy solution:
Credits can be defined as the parts of the deconstructed conversion - Fractions of one whole.
The number of fractions is proportional to the number of channels contributing to the conversion, while the sum should always be 1.
For example, if the conversion is a result of user interacting with 4 different channels (4 sessions), the selected model will split the conversion into the same number of fractions (4 fractions), each valuing between 0 and 1 (Yes, by some models, such as first touch or last touch, the channels can receive as low as 0 or as high as 1 credits). The resulting fractions can be something like:
The end result is an easy-to-consume table where channels are represented as dimensions and accumulated credits are metrics.
The core concept of attribution modelling consists of two major components:
This is a pretty simple, straightforward method and, on the surface level, removes the complexities associated with deep analytical processes.
The modern customer journey is not a simple process. It involves multiple channels and there’s always some form of co-dependency between them. The actual conversion cost is also spread across these channels.
The key to successful advertising performance analysis is on one hand, understanding this co-dependency and on the other - Keeping the tabs on the actual money you’ve spent in this process:
And most importantly - Where the freed up money can be invested into?
Does the attribution modelling answer those questions?
The simple answer is no.
The key problem with the attribution modelling is the core concept itself - Splitting up the conversion into multiple fractions to credit each channel separately, independent of others.
We have already explained how this method simplified the interaction with the data and report itself.
But with the simplification, we lost one of the most crucial pieces of information - The real place of each channel in the whole picture.
The advertising channel might be able to drive conversions independently, that’s true, but not always. Separate channels often affect and define each other’s role in the user conversion funnel.
The attribution modelling strips all information about this - By evaluating each channel out of the context, it reduces the complex channel relationships and your advertising budget, spent on creating these relationships, to just the list of channels and numbers out of the context.
Simply to say, an attribution model report showing that 101.75 conversions are attributed to Meta Ads and 43.35 are attributed to Google Ads does not ever mean that Meta Ads is responsible for 101.75 conversions while Google Ads is just for 43.25 ones.
Seeing 101.75 as a number representing conversions does not even mean that Meta Ads is truly responsible for at least 101 whole conversions, instead it could be 0.5 credits for 202 conversions or 0.7 credits for 144 conversions.
This might sound confusing, but bear with us a little bit more. We’ll try to help you understand it better.
For this example, let’s say the business is running ads on Meta Ads and Google Ads. They have already spent $30 on Meta and $90 on Google which lead to one conversion.
The converted user took three sessions from different channels until making the conversion.
Now let’s see how different attribution models would report the converted user journey and performance of each channel.
In the meantime what really happened is that the user had a journey with the business that included multiple interactions from different channels until it ended with the conversion:
The actual conversion cost was $120 - A conversion journey included both channels and the whole $120 was spent to convert this user.
None of the attribution models give advertisers the actual view of how or why the user converted and what it cost the business.
While attribution modelling conceptually is simple, it still brings its own complexities.
Specifically, multiple models that are not intuitive enough to navigate through them without uncertainty and question - Does this model fit my specific case?
We have already mentioned the most common models:
The usual way to overcome the overwhelm of choosing the right model was to place different models side-by-side and compare the results.
But this could also lead the advertiser to uncertainty how to properly interpret the numbers into insights and then into actions.
You may have heard that evaluating several models gives you the idea about different aspects of the channels, such as identifying the ones working best as the conversion starter or closer.
We have to tell you it’s quite challenging to do that and the reason for this brings us to the next issue with the attribution modelling.
There’s a deeper issue with the models that somehow has been sidelined for a long time -
Even if we take attribution modelling as it is, the answer is still “sometimes” at best and the longer the user conversion lasts, more the answer shifts to “no”.
And it’s no coincidence that Google deprecated all attribution models except last click and data-driven.
According to Google:
"Note: The first click, linear, time decay, and position-based attribution models are no longer supported by Google"
To better understand the issue, let’s focus on the models such as First click and Position-based.
How are these models usually defined? Both of these models emphasize the first channel, first point of user interaction. There’s no doubt that it’s important to understand what works best to get users' attention (even though, position based model doesn’t really focus on it with the numbers)
But does the first touch in the attribution model really represent the first user interaction?
NO.
In reality, the sessions, including the very first session, are all relative to the selected conversion window.
Let's examine the following scenario:
This is a rough example, but the idea is that all reports and all models you see are always relative to the conversion window selected.
We don’t say that the conversion window is the problem - In reality, it’s actually a necessary tool to define the clear time frame as it’s impossible to reach the conclusive results without taking the time into account.
Rather, our goal is to show you how some of the attribution models can become a collection of confusing numbers really fast.
To be clear, this article is not about attribution modelling being an unusable technique that has no place in modern digital marketing.
In reality, attribution modelling still has its use, but it’s important to know what it really should be useful for.
As already stated above, each attribution model calculates one number (A sum of all fractions credited from each conversion) per each channel, independent of all other factors. All other metrics, including cost per result, is the derivative of this number.
Unlike advertisers, who need to understand the customer journey and the actual costs and returns, advertising platforms need signals in their simplest forms to evaluate the performance of each campaign, ad group / ad set and ad.
By default, these ads are independent from each other in the same way as they are represented in the attribution modelling reports.
All the context they have, comes from the advertisers themselves - how they planned the campaign structure and conversion funnel, what they intended with each creative and copy and what they are expecting from their advertising efforts - This is all human.
Meta or Google Ads ad distribution system does not need context to target the audience or distribute ads - They simply need the numbers to evaluate the ad quality and adjust the distribution.
Attribution modelling can bring this quantitative information in its simplest form without any requirement to understand the context of the ad or the customer journey.
Attributing more conversions signals the advertising platform that the ad is engaging with the audience and it can be delivered further.
Even automatic optimizations, such as advantage campaign budget on Meta solely relies on the conversion numbers attributed to each ad.
Today attribution modelling has its own place in the digital marketing ecosystem, there’s no denying in that.
The problem with attribution modelling is not that it’s used, but rather it is used incorrectly. It is the purely quantitative information in its simplest form that demands advertisers to make decisions in the same way as advertising platforms. But advertising platforms are already making decisions like advertising platforms.
Advertisers' job, while analyzing the ad performance, is not just to evaluate if the ad is responding to the audience or not.
If you have thought about it, know that you are not alone. We are constantly thinking about it too.
There are ways - new approaches across the market, including our own brand-new approach with advertising performance measurement.
You can check it out on our website or contact us to find out more about it.
