L'Oréal Luxe achieved 57% sales growth for Lancôme's D2C website. How? Read case study to find out
All articles
Articles
Marketing attribution is not what you probably think

Marketing attribution is not what you probably think

Marketing attribution is a crucial concept for every business, as it helps to understand which marketing efforts drive results and which do not. But the definition of marketing attribution is still confusing for some marketers.
Marketing attribution is not what you probably think Pavel Petrinich
Articles
Marketing attribution is not what you probably think
14 min read

Marketing attribution is a crucial concept for every business, as it helps to understand which marketing efforts drive results and which do not.

By analysing attribution data, marketers can make informed budget allocation decisions to achieve the highest return on their advertising investments. At the same time, advertising platforms such as Google Ads, Facebook Ads, and others use attribution data as feedback to improve their targeting and bidding algorithms. 

Despite being such an important part of every business, attribution can still be a challenge for many marketers. The primary reason is often a misunderstanding of what marketing attribution actually is.

This article aims to resolve a typical misconception about what attribution modelling is and provide guidance on how to think about marketing attribution correctly.

What exactly is marketing attribution?

It is common to think that marketing attribution is the process of assigning credit from a conversion to the marketing channels and campaigns that led to that conversion. 

However, that is not entirely true. This misconception often leads to wrong decisions and prevents businesses from achieving their primary goal, which is understanding the true value of their marketing traffic sources.

Here is a correct definition of marketing attribution, provided by Wikipedia

Marketing attribution true definition

But what is the difference, you may ask? To answer this question, let’s first explore how marketing attribution has evolved over time.

For a long time, marketing attribution models operated in the same way: 

  1. A conversion happens. It could be a placed order on an e-commerce website, a website lead, a hotel booking, a phone call, or a closed deal — this varies from business to business. Let’s say the conversion is a purchase for $300.
  2. The user’s path to conversion is analysed. For example, before converting, the user interacted with the website three times: initially, they visited the website from a Google Ads campaign; then they returned from a retargeting display campaign; and finally, they converted from an email newsletter.
  3. Depending on the chosen attribution model, the business decides how to split the conversion value between these three touchpoints. For example, using the Last Click or last non-direct click attribution models, the entire value of the conversion ($300) would be assigned to the email channel.

But is it fair? 

This is where the marketing attribution battles begin.

Should email really receive all the credit? It doesn’t seem fair.

— Let’s give all the credit to Google Ads because it was the first channel!

Hmm, that doesn’t sound fair either.

— What about giving each touchpoint equal credit then? $100 to Email, $100 to Retargeting, $100 to Google Ads?

But why do we give equal credit? Maybe one channel was more valuable than the others.

— But what exactly does “more valuable” mean? How do we quantify this value?

Let’s use algorithms!

In trying to find the right answer, marketers often go down the attribution rabbit hole, only to find out that no matter which model they choose, they are still not happy with the result.

In this search for the “holy grail,” some businesses have invested millions of dollars trying to build the best attribution model, only to find themselves returning to the good old Last Click model.

Why?

Well, the primary reason is that there was no significant difference in the results. What’s the point in using some complex attribution logic (That might even be hard to explain!) if the final marketing report still shows similar data to the simple Last Click model?

At this stage, some attribution experts start to give confident talks at conferences about how attribution doesn’t work anymore without even knowing what marketing attribution really means in the first place.

The biggest misconception of marketing attribution

Let’s recap the correct definition of marketing attribution provided by Wikipedia:

“The identification of a set of user actions that contribute to a desired outcome, and then the assignment of a value to each of these events.”

Now, let’s compare it with the common and misleading definition of attribution:

“The process of assigning credit from a conversion to the marketing channels and campaigns that led to this conversion.”

If you didn’t notice the difference, here’s a hint:

The biggest misconception of marketing attribution lies in thinking that a traffic source should receive value from a single conversion it contributed to.

This is the single biggest reason why many marketing attribution models fail to deliver. By trying to distribute the value from a single conversion, marketers and analysts become trapped in the idea of tracking customer journeys. 

They want to know which traffic sources or touchpoints a user interacted with before converting and then distribute the value from this conversion accordingly. However, this approach can only work if you track the full customer journey from the very first touchpoint to the very last. 

If the customer journey is not complete, then this whole approach falls apart. This is precisely what happens in most cases. No matter which attribution model is used, whether it is a simple one-touch model or a sophisticated data-driven model, neither can assign any value to traffic sources they simply can’t see.

It is not possible to track users’ real path to a conversion

One of the primary reasons — are cross-device and cross-browser interactions. 

Today’s consumers often use multiple devices and browsers to access the web, making it challenging to track their activity across different platforms. This fragmentation makes it difficult to get a complete picture of a customer’s journey to conversion.

With the rise of in-app browsers of popular mobile apps, such as Facebook, Instagram, and Pinterest, this became an even more serious problem, as these in-app browsers contain a unique set of cookies. So in case, the person switches to their primary browser such as Safari or Chrome, the new customer journey will start, while the previous will end without a conversion. 

The next challenge is the increasing prevalence of cookie restrictions. 

Cookies are small pieces of data that are used to remember the user on the same website and track their activity across the web. However, many browsers are now blocking or deleting cookies, making it much more difficult to track online behaviour.

For example, Intelligent Tracking Prevention in Safari not only blocks third-party cookies, but also limits the lifespan of first-party cookies to 7 days, and in some cases, just to 1 day. 

Browsers delete cookies to provide security online

Short attribution windows are also a problem. 

While the customer journey may be long and complex, the attribution window is limited. For example, the default attribution window of Facebook Ad Manager is a 7-day post click. This means that unless conversion happens within 7 days after the click, it will not be observed at all. This creates serious challenges for ad campaign optimisation, as conversion signals are an important feedback mechanism for ad targeting algorithms to learn. 

Private browsing is another issue that affects marketing attribution. 

Many users now use private browsing modes, which prevents cookies from being stored on their devices. This makes it impossible to track their activity across the web, making it more challenging to attribute conversions accurately.

As a result, marketing traffic sources do not receive fair value.

Marketing traffic sources do not receive fair valueInitial traffic sources will not receive any credit for the conversion they contributed to.

Due to the challenges described above, the majority of website visits do not receive any value, even if they significantly contribute to conversions and revenue. In fact, up to 90% of conversions are not attributed to the initial traffic sources.

This creates two main problems for marketing teams:

1. Correct measurement and budget allocation are not possible due to the inability to track customer journeys. 

All traditional marketing attribution models that distribute the value from a single conversion will fail to deliver. Even with First Click attribution, most upper-funnel campaigns are significantly undervalued.

This is the reason why most of the time marketers do not see much difference when comparing First Click attribution reports with the Last Click ones. In the image above, both with the First and Last Click, the entire value of the conversion will be attributed to a single traffic source - Direct/None.

2. Ad platforms struggle to target and optimise campaigns effectively.

Conversions are essential feedback mechanisms for Google and Facebook Ads, and other ad platforms. They rely on conversions to train their smart bidding algorithms and improve their targeting and optimisation capabilities. Providing ad platforms with more value signals as soon as possible is crucial for effective ad buying.

The problem is that due to broken customer journeys, the majority of clicks do not receive any value, as users rarely convert within a single website visit or cookie. It is especially challenging for businesses promoting high-value products or services, where buying journeys are longer and more complex. As a result, a lack of conversion signals prevents ad platforms from delivering desired results.

Due to incomplete and delayed feedback about the true value of each click, ad platforms such as Google and Facebook Ads are unable to deliver their best.

How to think about marketing attribution differently

If you have been asking yourself, “How to distribute the value from the conversion to the traffic sources that led to this conversion?”, you should start asking a different question.

The correct question is this: “How to distribute the total conversion value between the traffic sources according to their contribution to driving this value?”

By distributing the value from the total number of conversions, rather than a single conversion, you will guarantee that each traffic source gets a proper value, even if the actual conversion does not happen within the same cookie. For example:

  • The advertising campaign generated a website visit on mobile.
  • The user did not convert during the initial website visit.
  • After some time, the user returns to the website from the desktop and makes a purchase.

Here, the initial advertising campaign is responsible for driving a sale, even though it will not be directly attributed to the campaign by any attribution model that relies on tracking the user’s path to conversion using cookies. However, it is clear that both website visits contributed to the purchase and, as a result, should both receive value.

The question now is: How to measure this contribution?

How to measure the contribution of each traffic source

People cannot make a purchase without visiting your website and going through a set of steps to complete the conversion. 

Before buying, users will do their research, visit multiple pages on the website, view product images, read FAQs, and take other actions before making a final purchasing decision. Therefore, every website visit and every action taken has an incremental effect on driving a user towards a goal — whether it is a purchase, booking, or lead form completion.

By analysing user behaviour patterns, you can identify which set of actions lead to the conversion, and which don’t. This data, when analysed properly, will provide you with valuable insights into how different website visits contribute to sales.

Here is a simple example:

  • The first user visited your website but left after a few seconds without converting. 
  • At the same time, there might be another user who also left the website without purchasing, but unlike the first one, they’ve spent 15 minutes browsing through the website, carefully choosing the product they are interested in, and returning to it multiple times.

Are these two non-converting website visits equal?

Of course not. The chances that the second user will complete a purchase are much higher than in the first case. 

However, what you will see in your analytics tool is that two non-converting visits have the same value, as they did not result in directly attributed conversions. Ad platforms that drove these users to the website will also receive the same information and will be unable to target more high-quality users with a higher likelihood to purchase.

Now, imagine that you could assign value to these two non-converting visits depending on the information you have. Most likely, you will assign far more value to a more engaged website visit. Even if you assign a value using synthetic metrics (i.e., 1 point to ‘Visit A’ and 5 points to ‘Visit B’), it will still help both you and your ad platforms understand which traffic sources drive more valuable users to your website with the highest potential to convert in the future.

However, this approach is not entirely reliable, especially at scale. The primary problem is that real user behaviour patterns are much more complex than in our example above, and they cannot be described using a simple rule-based formula. 

Even if the user spent at least 2 minutes on the website, it doesn’t necessarily mean that they are actually moving towards the purchase. Many additional factors and nuances need to be considered, which the human brain cannot process manually. These factors include information about the user, the context of the website visit, product pricing if you are selling multiple products, and many other data points.

As a result, when you have thousands of website visits, you cannot make correct conclusions about how valuable each visit is using a gut feeling that this particular website visit is more valuable than another. 

To find correlations between the enormous amount of information and actual conversions, it can be helpful to use the power of cloud computing and Artificial Intelligence.

How to assign proper value to each visit using Machine Learning 

At SegmentStream, Machine Learning is a core technology of our Conversion Modelling Platform. We use a combination of 1st-party website data with our proprietary Machine Learning algorithms to measure the incremental contribution of each website visit to future conversions and assign a proper monetary value to it. 

By relying on AI, we guarantee that each website visit is being accurately measured, and the value is calculated in a truly data-driven way. 

Let’s review how we do this at SegmentStream step by step.

Step 1: Each website visit is immediately analysed using Machine Learning

SegmentStream has two Machine Learning models to predict the user’s conversion probability and calculate the predicted conversion value. 

To make these predictions, SegmentStream trains an ML model based on a historical dataset of available user behaviour and other 1st-party data, which is unique for each website.

Once the ML model is trained and the accuracy of predictions is validated, we start measuring the incremental value of each website visit.

Step 2: Each website visit gets value according to its incremental contribution to future conversions

To calculate the incremental contribution to future conversions, first, we analyse the shift in conversion probability since the last website visit (if it’s known). 

For example, if a user visited the website and the probability to buy reached 15%, but then left without converting, the visit will be responsible for a 15% shift in conversion probability if this was the first website visit within the analysed cookie. 

However, if the user visited the website previously, let’s say, an hour ago, and reached a 10% probability to convert there, the following website visit will be responsible for a 5% shift in conversion probability. 

This way, SegmentStream measures the incremental impact of each website visit on future conversions, and not just a conversion probability for each website visit.

Measuring the incremental impact in SegmentStream

To understand the value of the website visit in monetary terms, we multiply this incremental impact (%) by the predicted conversion value. As a result, each website visit will get an accurate monetary value even if the actual conversion did not happen yet.

Step 3: Creating Modelled Conversions for valuable visits

Of course, not all website visits have a high impact on driving future conversions.

To ensure the quality of reporting and to minimise noise, it is recommended to exclude non-valuable visits from analysis, where the shift in conversion probability was minimal, i.e. less than 1-2%.

On the other hand, if the contribution to a future conversion was significant (higher than the minimum threshold), SegmentStream will create a Modelled Conversion, which will be immediately assigned to the traffic source that generated this particular visit. 

By analysing how many Modelled Conversions each traffic source has generated, you will be able to understand how many valuable website visitors they have generated for your business.

For example, if one traffic source drove 2x more Modelled Conversions than another, this essentially means that by investing in the first traffic source, you can generate two times more website visits where users have a high likelihood to convert in the future, even if the actual conversion will not be observed due to cookie tracking limitations.

Each traffic source finally gets the value it deserves

Unlike traditional attribution models that assign a value from an actual conversion retrospectively, using Conversion Modelling guarantees that each traffic source gets the value according to its incremental contribution in driving future sales, even if the actual conversion happens from another device, browser, or cookie. 

Another advantage of this approach is that the traffic source receives the value immediately, which is especially important for ad platforms’ algorithms to learn faster.

Each traffic source finally gets the value it deserves Each website visit will immediately get a value even if the actual conversion hasn’t happened yet or happened in another cookie.

Understanding the true ROAS and CPA of each traffic source

By aggregating Conversion Modelling data across all website visits, we can see which traffic sources drove the most incremental value to the business in a form of Modelled Conversions:

  • Facebook Ads - 250 Modelled Conversions, $18,000 in Modelled Value
  • Google Ads - 300 Modelled Conversions, $10,000 in Modelled Value
  • Direct/None - 20 Modelled Conversions, $5,000 in Modelled Value

In total, we have 570 Modelled Conversions with $33,000 in Modelled Value.

By understanding the share of traffic sources in driving total Modelled Value, we can identify the contribution of each marketing channel or campaign in driving incremental value to the business in percentages. For example, in our particular example, Facebook Ads have an impact of 55% ($18,000 / $33,000).

This way we have identified the contribution of each marketing channel, and with this data, we can finally answer our marketing attribution question: “How to distribute the total conversion value between the traffic sources according to their contribution in driving this value?”

So, the last step we need to take to define how many real conversions and revenue the channel is responsible for is to multiply this contribution (%) by the total amount of actual revenue received across all traffic sources. For example, if the business has generated $40,000 in total sales, Facebook is responsible for 55% of this value. 

So, if previously your analytics tools reported, let’s say, just $7,000 in directly attributed conversions for Facebook Ads, now you can find out the true sales impact of this channel which is $22,000. 

With this in mind, by distributing the total conversions value between the traffic sources according to their relative share in generating Modelled Conversions, you finally understand the true value of all your paid channels and campaigns and calculate important metrics such as ROAS or CPA of marketing traffic source. 

This, in turn, will help you make correct and confident budget allocation decisions and improve your total Marketing Mix. 

At the same time, by sending the information about the value of each click back to ad platforms such as Google and Facebook Ads, you will significantly improve their performance. In fact, you can expect to send up to 20x more quality signals about the value of each ad click, which will result in enhanced ad targeting, faster learning, and improved campaign optimisation.

If you would like to learn more about why Conversion Modelling is the easiest way to improve the performance of your Google Ads and Facebook Ads campaigns, check out this article.

To sum up

Attribution, despite being one of the most important things in digital marketing, is often misunderstood. Relying on the common knowledge that digital marketing attribution is responsible for assigning a value from a single conversion to the traffic sources that led to this conversion creates many challenges in understanding the true impact of digital marketing sources. 

Marketing sources that contribute to conversions and revenue often simply do not get proper credit due to tracking restrictions and complex customer journeys. Therefore, any model that relies on tracking a user’s path to conversion will fail to deliver accurate insights and will be misleading for the marketing teams in their budget reallocation process.

This creates the same challenge for popular ad platforms as well since they rely on accurate attribution data to train and tune their automated targeting and bidding algorithms. 

Therefore, it is crucial to shift the focus from a single conversion to the total conversions that the business has generated across all traffic sources combined. Then, this value should be distributed between traffic sources accordingly to their incremental contribution to driving sales, even if these sales cannot be attributed directly.

There are multiple ways to achieve this. However, one of the easiest and most reliable ways to understand the contribution of each website visit is Conversion Modelling. 

This is an innovative measurement approach that relies on using a combination of first-party data and Machine Learning to assign value to each website visit based on its impact on driving predicted conversions, rather than past conversions. 

Conversion Modelling eliminates all possible problems of the traditional, retrospective attribution models as it guarantees that each website visit and traffic source will get the accurate value even if the real conversion happens from another device, browser, or cookie.

You can use Conversion Modelling to gain visibility into the real value of your traffic sources, as well as unlock the full potential of your ad campaigns by providing them with immediate feedback about the true incremental value of each paid click. 

To learn more about Conversion Modelling and its capabilities, visit the SegmentStream website and book a demo with one of our solution consultants.

See SegmentStream in action
Find out the true value of your website traffic sources and boost online ads' performance
Thank you!
We'll contact you soon!

You might also be interested in

More articles

Get started with SegmentStream

Learn about Conversion Modelling and why it is a true next-generation solution to outdated marketing attribution and conversion tracking tools.