Attribution Modeling: The Future of Marketing

Over the last few years, the focus in marketing has been which online channel or medium is best for displaying advertising. Today, it is increasingly important to be present in as many advertising channels as possible. This is where attribution modeling comes in - being able to assign a conversion to more than one channel.

The idea of being present on many different channels is dismissed by some as too "watered down", but it is a common practice in many companies. It is not enough to just be present on facebook, or affiliate networks for example (depending on the brand). Several channels should be combined.

The problem here is with the evaluation of this approach - it is no longer up to date. With many web analysis tools, such as Google Analytics, it is only possible to evaluate measures such as Facebook or Google Ads using the "Last Click" method. This is not correct - each channel stands for a different characteristic - the key words being "push" and "pull" marketing. For example, when a user searches with a long tail request in Google, their intent to purchase is clear. This is more likely to lead to a conversion than a displayed advert. However, the customer might never have searched for a product if they hadn't seen the advertisement displayed. In order to be able to assess each channel accordingly, the "attribution model" can be applied. The objective here is to determine the efficiency of individual channels or measures.

Similar to a soccer team, the striker would have difficulty scoring if there was no defender, goalkeeper, coach, sponsor, etc. All of these players and organizers are essential for scoring goals. In traditional advertising planning, this is often ignored, or too much attention is paid to the more expensive channels such as TV, radio, or print.

The attribution model can show which budgets are best used in which channel, and how these channels interact with each other.

Attribution – What is it?

In computer science, attribute mapping means the assignment of values to the attributes of an object. In mathematics, attributes are used to represent certain settings that a system assumes in relation to formulas or inference rules. Put simply, attribution is when several small pieces of information are collected to generate the "main information." Translated into conversions in online marketing, this means that a conversion is made up of many different "smaller" conversions. In addition, each of these "individual" smaller conversions has a different value.

Machine learning and the entire topic of AI is also based on these models. This alone makes the topic of attribution modelling interesting - it is not only relevant for the (online) marketing, but also for other technological areas.

Various marketing tools and advertising platforms, such as Google AdWords, BING, or Adobe Analytics, already offer the option of modifying the so-called "attribution." Various models are possible here, such as a position-based model, a linear attribution, or a data-driven model.

Google and Attribution

For some time now, Google has also been paying close attention to the topic of attribution. In some Google marketing tools, it is possible to intervene in this "system", or to modify the settings accordingly. Google Analytics and Google Ads are based on the so-called "Last Click" or "First Click" model. This means that the conversion is always assigned to the last (online) channel or vice versa.

In principle, this approach is correct. The problem only arises with a closer analysis: if a customer converts due to a display ad, but became aware of the product via the organic search, the conversion will be assigned to CPC (SEA), not to organic. Now, it would be interesting to know how valuable and important the SEA channel in this customer journey actually is.

• Last interaction: Conversion is attributed entirely to the last interaction

• First interaction: Conversion is attributed entirely to the first interaction

• Over time: The interaction that is closest to the time of conversion is given the larger conversion percentage. These are ideal for short campaigns, since they have a high conversion rate.

• Linear: The different interactions will share the contributions to a conversion. This is ideal for a uniform analysis.

• Position-based: A mixture of the First and Last Click models. The first and last interaction is awarded the highest percentage. This is ideal if increasing brand awareness is important, but also for a conversion.

• Data-driven: Google or a corresponding system will split the conversions or interactions itself. For this, it is important to possess a sufficiently large data pool, such as conversions.

Google Analytics

In Google Analytics, there are several ways to view achieved conversions and transactions without looking at the "Last Click" approach. One of these would be the insight offered by the multi-channel funnel under "multi-channel funnels" - "top conversion paths." This allows you to analyze how much the individual sources have contributed to conversions and the achievement of goals. Furthermore, you can see whether certain sources contribute more to direct purchases, or are more of a preparatory nature, or whether certain sources directly lead to a purchase. Marketing measures or budgets can then be adjusted accordingly.

Figure 1: Multi-Channel Funnel View in Google Analytics

Furthermore, it would be possible to create your own conversion segments or to perform additional analyses for various insights via secondary parameters.

Figure 2: Multi-Channel Segments, etc.

Essentially, the multi-channel funnel evaluation is a type of attribution evaluation. An analysis which shows the attribution of the respective advertising channel in more detail can be viewed under "Conversions" - "Attribution" - "Model Comparison Tool." Here you can see the influence of the different channels on the conversion.

Figure 3: Attribution tool overview and setting options

It is possible to compare the different models to see the extent to which a model change to a channel has an impact. For example, it would be an important insight to know that with the "last click" model, considerably more conversions are attributed to the "first interaction" than to the "last interaction."

In order to be able to view conversions and conversion values as quickly as possible, you should change this to "Conversion and Value" in the drop-down menu as part of one of the first steps of the analysis.

Figure 4: Conversion Value Selection

Furthermore, it is also possible to create your own mapping models or to include conversion segments. If none of the pre-selected mapping models meets the respective requirements, you can create a user-defined model for yourself.

Figure 5: Create your own attribution model

AdWords Attribution Models

In AdWords, the problem of mapping or attribution exists "twice:" when creating a conversion with the basic settings, this is always applied to the last clicked campaign ("Last Click"). However, this has essentially nothing to do with the "last click" model in Analytics. Google Ads (with its default settings) keeps the "First Click" model. The decisive difference here is that Google Ads internally calculates the campaign metrics using "LastClick."

We recommend resetting the Conversion Counting Model in this case to change the analysis and resulting settings. Such a change should have a positive effect on the entire campaign management.

When uncertain about this conversion, it is also possible to see in advance how a new attribution modeling setting would affect conversion distribution. This can be done in AdWords under "Tools" - "Click Analysis" - "Attribute Modeling."

Figure 6: Attribution in Google AdWords Preview

What if there are suddenly fewer conversions and the numbers plummet?

This is definitely not due to the settings of the attribution. The worst thing that can happen is that nothing will happen. It might sound a bit dull, but that’s the way it is.

It should have the positive effect that the conversions are "re-allocated" to the respective campaigns, meaning that the click prices can be customized individually. In this context, campaigns that do not generate conversions using the "last click" model, but which are very important in the "push & pull" category could become even more important. Therefore, all campaigns can be considered, not only the converting campaigns. It can also be determined if there is any causality there.

The system also updates itself regularly, which means the dynamic algorithms that the data-driven model contains always provide the right model for the customer, or the system selects the right model.

For the data-driven variation, a certain number of conversions must already be available, otherwise this option cannot be selected. Nevertheless, it is recommended to use the position-based model and not use the standardized first click data model.

Google Attribution Tool

One tool that will be released in the upcoming months is Google - Attribution. This is already being beta tested with selected customers and has received good feedback so far. With this tool, it will be possible to redistribute the respective marketing budget. Not only the Google internal marketing tools are shown here, but also social media, organic, etc. Furthermore, the tool is also cross-channel based, similar to Google Analytics.
This means that everything that can be represented in Google Analytics can also be represented in Attribution. In addition, it is also cross-device, which distinguishes the tool from many other providers in this field. That means Attribution can predict with a high degree of probability whether a user has cross-device activity. There will be an interface for returning to GoogleAds - this means it will be possible to be able to control Google Ads in such a way that most impressions or conversions are generated using the aggregated data from Attribution. This also allows you to set click bids, etc. more precisely.

Figure 7: Screenshot of Google Attribution Tool (Source: Search Engine Land)

More Tools and Alternatives

Of course, Google is not the only company to have taken up the subject of attribution more intensively - other system providers have also been dealing with it for some time. For example, the Adobe Analytics Cloud can also be used to analyze the sequence of events that lead to a conversion in detail.

This is also where "customer contact points" such as call center phone calls can be integrated, or even consultation calls in an offline store, for example, if a customer card is available. Various anomaly detections and load analyses also provide the opportunity to personalize the product to address the customer in an even more specific way.

With Mapping & Attribution Modeling, insight into a uniform customer journey is also possible via the Webtrekk web analysis tool.

Figure 8: Webtrekk attribute settings during target creation (Source: Webtrekk)

Of course, there are other providers on the market like Kochava, Tune, Affise for Affiliate Marketing, Econda, etc. that allow you to take a closer look at contact chains. SEO tools such as Ryte, Searchmetrics, or Xovi also offer ways and means of integrating tools such as Adobe Analytics, Google Analytics, the Search Console, or Webtrekk.

Conclusion

Irrespective of which system you decide on, or how you deal with the issue of attribute modeling: without a (technical) check to see if the respective system is correctly integrated (fires the Javascript snippet, is the Tag Manager correctly implemented, is the SDK integrated without errors etc.) and all settings have been made correctly (keyword channel grouping Google Analytics or wt-parameter names in Webtrekk), you should not jump to any preliminary conclusions and it is better to be on the safe side again. The same applies here as everywhere else.

In any case, it is almost certain that attribution modeling will be the working method of the future for online marketing. By understanding a few basics, it is also easy to understand in its entirety.