We live in a world of explosive e-commerce growth. Selling products and running ads online has become not only the norm but has reached a scale and precision previously unimaginable by offline retailers. 

For a traditional offline retailer, it is difficult to both collect and use data to accurately measure the effectiveness and result of any ads they are running, whether that’s billboard, newspaper ads, etc. 

However, online advertising has opened a door for more precise measurements to occur. 

If you are already using one of the online channels to promote your brand or products (such as Facebook, Google, Bing), then you can definitely view relevant data related to each of those channels with ease. 

However, the abundance of multi-channel data created new unique challenges for modern advertisers, originating from one central issue - all advertising channels want to take sole credit of conversion, whereas in reality credits should be shared across different channels.  

Any solution that attempts to solve this “attribution” problem by attempting to create a rule or a set of rules to accurately determine and assign credits to different paid and unpaid channels are called an attribution model.

The purpose of this article is to walk you through examples of different attribution models provided by default by Google products (this is both Google Analytics and Google Ads), and offer you some clear suggestions on which models you should pick.

While we are only talking about Google attribution models here, the knowledge explained in this article can very well transition to other channels such as Facebook and beyond, offering even more value for your understanding of attribution.

We will begin by introducing the CASE that we will use throughout this article:

1. Customer A clicks on a Google ad and visits your website.

2. She entered the site again via Google search and signed up for your newsletter.

3. Two weeks later, she clicked through to a blog post via the newsletter.

4. A month later, she returned to your website and visited your product and made a purchase.

Now let’s look at who should get credits for each of the attribution models.

Last-Click – The Default Model

Let’s begin with the simplest, and default attribution channel for both Google Ads and Google Analytics – last click. 

Last click gives all credit of the conversion to the very last channel (ad or not) and corresponding keyword if applies. 

Returning to our case, here is the distribution of credit for the Last-click Model.

Direct gets 100% credit

In Last-click Model, your Google analytics report will only show Direct and give 100% credit, while others such as Google ads or Google organic or Email will not be shown.

The Last click almost discounts the role of all other channels on the conversion path, so it is not rigorous enough to get an accurate result. 

One thing that I do want to stop and note an important distinction between Google Analytics vs Google Ads.

Google Analytics’ advertising channel considers all possible paid and unpaid channels when computing its attribution model, whereas Google Ads only considers ads that are within its own platform (so Google Ads gets 1 credit ).

The last click model gives all the credit to the last channel, so the main problem is that it ignores the first click and the other channels in the complete conversion path. In fact, the consumer's first click is also important to fully show the user which channels play a role in attracting new users.

Next, let's discuss the first-click.

First Click – The Basic Impression Model

If your company is running a brand awareness advertising campaign, consumers’ first click is actually very important to you.

First Click Model gives all credit for the conversion to the first-clicked channel and corresponding keyword.

Going back to our cases again, here is the distribution of credit for the First-click Model.

Google ads gets 100% credit

The First Click Model and the Last Click Model are single-touch models, so the problems are similar. 

The first click model attributes all credit to the consumer’s first click but ignores the last click and other channels on the conversion path. Therefore, it still cannot truly reflect all the channels in the customer conversion process.

To solve this single-touch model problem, we will look at the other three models, Linear Model, Time-Decay Model and Position-based Model.

Next, we move on to the Linear Model.

Linear Model-  Equal Credit Model 

Linear Model distributes the credit for the conversion equally across all channels on the conversion path.

Here is the distribution of credit for the Linear Model.

Google ad: 25% credit

Google organic: 25% credit

Email: 25% credit

Direct: 25% credit

In the linear attribution model, each channel on the path will receive equal credit. There are four channels on the conversion path in the above case, and each channel will receive 25% credit. 

Unlike First Click and Last Click, Linear Model can ensure that every channel along the path gets the same conversion credit.

However, many people think that credit should not be evenly distributed because each channel has a different contribution. How to judge which channel has played more value in conversion?

For this question, two other better attribution models can help you. 

A)Time Decay Model  B)Position-based Model

Time decay model -> The adjusted last-touch

Time decay gives more credit to interactions that happened closer in time to conversions.

In other words, the later in the customer journey a channel is, the more credits it would be assigned. 

Here is the distribution of credit for the Time decay Model.

Google ad: less credit

Google organic: some credit

Email: more credit

Direct: most credit

The Time-decay Model gives the credit to the above four channels, but the latter two email campaigns and the direct search will receive a higher percentage of credit.

The specific algorithm behind computation for those credits is beyond the scope of this article, and generally not important for most use cases as long as you understand the concept. 

Similar to the linear model, both the time decay model and the linear model shows all channels on the conversion path, but the time decay model is more sophisticated in assigning credits than the Linear Model. 

If your customer has a long conversion cycle and does not have an extremely sophisticated advertising strategy, then time-decay is the best attribution model for you.

However, if your business is running a large number of complex campaigns and channels to attract new customers, or your main business is concentrated on top-of-funnel conversions, then this model will not be the best model for you, as your focus will be underweighted by this model. 

In that case, a position-based model or custom model is recommended. 

Position-based Model  (we recommend)

Position-based models give a large portion of the credit to both the first and last interactions, and then the remaining is allocated to evenly other interactions along the path.

Going back to our cases again, this is the credit assignment for all of those channels (assuming first and last gets 40% each):

Google ad: 40%

Google organic: 10%

Email: 10%

Direct: 40%

In the case of position-based model, 40% would be assigned to the Google ad and Direct. Then, the remaining 20% would be split across the google organic search and email.

It can be seen that the most significant difference between the time-decay model and the position-based model is the first interaction with the user.

In the Position-based Model, you can see which channels are most suitable for attracting interested new customers and which channels are suitable for sales conversions.

It also evenly distributes other channels along the conversion path.

Position-based Model should be used when you want to emphasize both the initial impression and the final conversion steps of your customers, which generally applies to companies that have a distinct split between prospecting (or top of the funnel) and retargeting (bottom of the funnel) activities. 

Data-Driven Model (we recommend)

Data-drive Model distributes credit for the conversions based on your past data for this conversion action. It's different from the other models in that it assigns a percentage of contributions based on the historical data of the touchpoints on the conversion path, so it is relatively fairer and more reasonable.

This model is generally very difficult to compute without a mature data science team but Google is offered for free behind a heavy data requirement (a certain amount of money spent + conversions). 

To use this model, you need a minimum of the following in 30 days:

3,000  ad interactions

300     conversions

Google has produced multiple studies showing that data-driven models in general make the best results in ad performance, so I highly recommend using it if you have access. 

Conclusion

In conclusion, here is a summary of the three models we recommend you use: 

Time Decay Model you can choose this model when the consumer’s first touch point is not that important.

Position-based Model: you can choose this model when both the user’s first touchpoint and the last touchpoint before conversion are essential to you.