Czym jest atrybucja? #AlfabetEcommerce

Advertising opportunities today are very broad. Consequently, firstly, the consumer decision path is becoming longer and increasingly chaotic, and secondly, we have more and more ways and channels to reach these consumers. The consumer path from a person’s first contact with a brand to the purchase is not linear — there is a complicated network of touchpoints that varies from person to person.

What is attribution?

Attribution helps us organize this consumer behavior. In a broad sense, attribution can be defined as assignment. In the marketing world, attribution is a rule or set of rules by which we assign conversion value to a channel or channels located on the aforementioned chaotic consumer purchase path.

Thanks to attribution, we can see how customers learn about and buy a product or service. The first step in this process is identifying the set of user actions, establishing the sequence of touchpoint channels, and then assigning a value to each of these actions. Attribution makes the advertising world less complicated.

With so many media types, channels, and user data, it can be difficult to determine where to invest. Attribution models allow us to determine where these efforts are most successful, helping to optimize the conversion path.

By creating an attribution model, one can determine the ROI associated with each marketing channel and check which channel should be assigned the most resources.

What attribution models do we distinguish? Single- and multi-touch, first- and last-click...

There are various attribution models we can rely on. They are selected depending on the specifics of the business, the final goal of the analysis, and the advertising campaign.

Models can be divided into those where the entire value is assigned to a single channel on the path (single-touch), and those where the value is divided among all channels depending on the assigned weights (multi-touch).

Before describing these models in more detail, it is worth noting that when speaking of digital advertising touchpoints, I am referring not only to clicks but also to impressions. Most often, in single-touch models, the entire value is given to the customer's first or last touchpoint with the ad.

These are the first interaction and last interaction models, respectively. In the first interaction model, restrictions on exactly what type of interaction we take into account are set less frequently. However, the second model can be limited in various ways. The last interaction can be:

• the consumer's actual last step (click or impression) before purchase;

• the consumer's last click that is not a direct entry to the brand's website;

• the last Google Ads advertisement the customer clicked before purchasing, i.e., the last Google Ads interaction.

Single-touch models are perfect for understanding how the customer journey begins, what their first contact with the brand is (first-click), or what prompts them to make the final decision and purchase (last-click). Among multi-touch models, the following can be distinguished:

• linear – credit is assigned equally to every channel on the path;

• position-based – assigning weight to value depending on the channel's position on the path. Typically, greater value is assigned to the first and last interaction, and the rest is divided equally among the remaining channels;

• time decay – greater values are assigned to channels located closer to the end of the path. The standard period after which the value drops is 7 days, meaning an action that happened 8 days ago will be assigned half the value;

• data-driven model – Machine Learning is used to assign value to each channel.

If we compare selected models, we will see that the number of conversions assigned to specific channels can differ significantly depending on the model. This shows that every channel on the path plays its own role. We can distinguish the following roles:

• opening – most often the first contact with the brand, attracts attention;

• assisting – supports other channels, maintains brand awareness;

• closing – most often these channels finalize the conversion, prompting the customer to purchase.

Comparing models helps understand every channel and how it supports the customer in learning about the brand and making decisions. Based on this data, we can create a fluid advertising and storytelling context.

Data-driven model

Let's stay with this model for a moment because it is talked about a lot, but not everyone understands what it is about and why it is so important.

The data-driven model actually answers the question: "how many conversions should actually be assigned to a given channel, taking into account its overall impact on the conversion path?". The data-driven model differs from others in that, firstly, Machine Learning is used for its calculation, and secondly, it takes into account not only converting paths but also non-converting ones.

Additionally, the model can consider factors such as time since conversion, number of interactions with the ad, ad display order, or device type. Calculating conversion in this model can be divided into two important stages:

• analysis of the actual state, the probability of converting from a specific channel;

• analysis of alternative paths by removing one channel and calculating how many conversions we lost in the process, i.e., calculating the so-called removal effect.

Tools necessary for creating a multi-channel attribution model

We want to see the fullest possible picture of our consumers' actions on the web. From organic clicks, through seeing display and video ads, to clicking on paid social ads. Unfortunately, we cannot connect all of this in all platforms, and currently—spoiler—there is no possibility to connect 100% of actions in one place at all—something will always be missing. What we must establish before starting data-driven attribution modeling:

• which channels we use,

• how detailed we want to compare them (per provider, per format, per keywords—this will help us correctly create our own channel grouping rather than using the standard one, which is most often very inaccurate),

• whether we want to track impressions separately,

• whether we need Organic and Direct in the paths.

Most often at Salestube, for attribution modeling, we use data from Campaign Manager or Google Analytics. Below I present a short set of information worth keeping in mind for each of these platforms.

Campaign Manager:

• connection of advertising activities from all Google platforms – Search Ads 360, Display & Video 360 – allows tracking both clicks and impressions (with appropriate prior technical settings on the tracking side);

• partial connection of Facebook data: thanks to CM we can track all ad clicks and impressions from campaigns that are based on behavioral data in Facebook (e.g., age, gender, location). We cannot track ad impressions from campaigns based on first-party data (e.g., remarketing) in CM;

• tracking Organic is possible but requires additional technical settings;

• custom channel grouping is available;

• data in CM is available for a maximum period of up to 2 months back;

• data can be exported to create your own custom analyses and use additional variables.

Google Analytics:

• standard GA does not capture channel impressions, only clicks themselves, but with Google Ads and Campaign Manager connected, it captures all ad clicks and impressions available there (so effectively, all info from CM simply transfers to GA). An important condition is that this cannot be the basic version of GA – only Google Analytics 360;

• includes Direct and Organic in conversion paths;

• custom channel grouping is available;

• data can be exported to create your own custom analyses and use additional variables.

Summary

When thinking about attribution, remember that:

• Attribution helps capture and understand the real value of channels in the chaotic middle of the consumer purchase path.

• Depending on the business and the goal of the analysis, we can use different rules for assigning value to channels.

• The data-driven attribution model uses advanced models and calculates the conversion probability of each channel using the removal effect.

• Standard attribution models help understand the role of each channel, while the data-driven attribution model helps fairly assign value to each of them and optimize the budget in the most optimal way.