Understanding Google Attribution Models: The Fundamentals

by Chris Wright

The process of attracting a potential client and closing a sale generally takes a variety of people, each with a unique contribution to the sales process. If you think of your digital marketing channels like a sales team, currently, most reporting provides all the accolades to the team member that happens to be present when the contract is signed. But what about the team that built the project plan and proposal and negotiated the terms of the deal? Or the person that made the initial contact with the lead at a conference? Or maybe you consider the critical sales steps to be done by the person who took that lead from the conference and nurtured the relationship and proved the unique value that your company can deliver.

Attribution modeling can help you identify who your high achieving contributors are by keeping the many touch points required to make a sale in perspective. In the above example, different models can help you see who your particularly effective sales people are even if they are not involved in generating leads or closing contracts. If you notice that Susan from the outbound sales team is super effective and making a huge difference to your bottom line you can go out and immediately invest in 20 more Susans (and fire Larry who’s been phoning it in for months).

The ultimate goal of attribution modeling is to provide perspective on the hard work that your converting and non-converting digital marketing channels are doing to help you succeed. This information will then help you make decisions on which of these channels are performing well, and merit more attention to the Susans and less attention to the Larrys. 

Google Analytics Attribution Models Explained

The Attribution Model Comparison Tool (in Google Analytics > Conversions >Attribution > Model Comparison Tool) has 7 attribution models to choose from. Each model assigns credit for conversions differently based on a set of rules. These rules are very simple (and quotable!), and based on when a marketing channel interacts with a user in the conversion path.

Ex: For the purposes of the explanations below let’s assume that Carol, a customer who loves your products, has engaged in the following activity online:

  • On Monday Carol gets a marketing email from you and visits the website by clicking the links in the email. She reads a blog post and leaves the site.
  • On Wednesday Carol uses a bookmark to visit your site again and browses a few items before leaving.
  • On Thursday Carol visits via a Paid Search link, and reads another blog post.
  • On Tuesday the following week, Carol visits again via her bookmark and finally makes a purchase.

Last Interaction

This model assigns credit to the last marketing channel to interact with a user before the conversion.

“If X closed the sale, X gets the credit.”

Carol’s last interaction was with her bookmark, so direct traffic would get credit under this model.

Last Non-Direct Click

This model is similar to Last Interaction in that it gives credit to the last marketing channel to interact with a user before conversion but it excludes Direct from this reckoning.

“If X closed the sale, X gets the credit … except Direct. Nobody likes that guy.”

Carol’s last interaction was Direct, but this model ignores direct touchpoints, so the credit would fall back to the Paid Search touch.

Last AdWords Click

This model provides the credit to the … you guessed it … last AdWords click.

“AdWords wins!”

Carol’s most recent AdWords click would get the credit, so if we assume her Paid Search touch came through Google, then that would get the credit. If that same touch came through Bing, however, direct traffic would retain credit.

First Interaction

This model gives all the credit to the first interaction with the user in the conversion path.

“I saw him first!”

Carol’s first touch was the marketing email, so 100% credit will go to email.


This model gives each interaction equal credit for the conversion.

“Give everyone that helped equal credit.”

The credit for Carol’s purchase would be distributed as follows: 25% credit is given to Paid Search and email, but 50% to Direct because there were two direct touchpoints.

Time Decay

This model gives the most credit to the interactions that are most recent and then progressively less credit the further away from the conversion the interaction is.

“Right now and close to right now is the most important.”

The default time frame for time-based adjustments is one week. So, Carol’s interactions would all receive equal credit, except the email. That results in ~28.57% being accredited to Paid Search, ~57.14% being accredited to Direct, and ~14.29% being accredited to Email (because it happened more than a week before the purchase and gets ½ the credit of the other touchpoints).

Position Based

This model gives 40% credit to the first and last interaction and then shares the remaining 20% credit among the remaining interactions.

“Right now and long, long ago.”

In our example Carol’s breakdown of credit would be 40% credit given to email, 50% credit given to Direct (40% + 10%), and 10% credit given to Paid Search.

*Bonus* The Data-Driven Attribution Model (Available to Analytics 360 Customers)

DDA uses data from your Google Analytics account to create an attribution model using the Shapley Value solution concept to make attribution recommendations. In very simple terms, it creates a model for determining how important a touchpoint was in the success of the conversion path by evaluating what would have happened if the interaction was absent.

Among our clients we are seeing an enthusiastic shift toward DDA. The data-driven model is a nice bonus for having an Analytics 360 license and doesn’t require any implementation or configuration work to set up. You simply toggle a switch in the Admin console and it is ready to start collecting data. If you are a 360 customer there is no reason not to enable it immediately and start building a customized attribution model.

In our ‘Carol’ example, there is no way of suggesting precisely how the DDA would distribute credit. It may find that one touchpoint has significantly higher impact on the propensity of a prospective customer to convert, but any distribution of credit among the channels is possible in theory. The good news is that it really depends on your business and your customers – it’s tailored to you!

Which Model is Right?

The biggest stumbling block that we have helped our clients through concerning attribution is shifting their understanding away from thinking of attribution as the default model (Last Non-Direct Click) and starting to evaluate the contributions of earlier funnel interactions to the success of the eventual conversion.

This shift in perspective beings with reviewing the model comparison tool in Google Analytics to help you see the real difference is between the models. The ideal way to use these attribution models is through comparison. The attribution reporting in Google Analytics is designed to let you see how these models differ in the way that they assign credit.

If you are accustomed to the Last Non-Direct Click model (the default way that Google Analytics attributes conversion) it is super interesting to see how credit is doled out when you compare it to the Linear model. This comparison might make you reconsider where you are putting your marketing dollars and where you are tasking your staff to focus.

Homework: Things You Should Try

Note: Before proceeding with any of these methods review which ‘conversions’ you have selected in the tool. Ideally you should be isolating one type of ‘conversion’ at a time.

  1. Open the Model Comparison Tool and compare First Interaction to Last Non-Direct Touch. Are you undervaluing one of your channels as an initiator? Try clicking on that channel name and see if you can identify an outlier campaign/email.
  2. Try the same with the Linear model, which channels are showing less and more value? Are there specific campaigns?
  3. Try the above exercises again…is there a significant difference in conversion value and total conversions when comparing these models? Does that imply one of your high/low conversion rate channels has low/high average over values?

Need some help? Don’t hesitate to give us a shout!

Chris Wright


As a Web Analyst at Napkyn Analytics, Chris uses digital data and visualization to tell the story of our clients' performance, helping them make decisions to move from insight to action.

See more posts from Chris