No, Google Analytics' default attribution model is not last click

Have you ever heard that Google Analytics' default attribution model is last click?

Well, that up there is a common idea, but, the reality is that it is not quite true. The default attribution model used in Google Analytics reports is not a last click one. Instead, its attribution model is called last non-direct click.

In my most recent poll here in LinkedIn, I asked the following question:

What is the attribution model used by default in most Google Analytics reports (except for the MCF ones)?

You will see the possible answers and next to it the percentage of respondents that clicked on that answer.

It was rather interesting to see that almost 70% of the people who answered the survey said that the model was last click.

Even though at a first glance one may think that a single word cannot make a significant difference, the reality is that if we compare last click with last non-direct click we will see that this last one generates an important over allocation of transactions and revenue that favors paid channels.

There are two main risks that can be derived from these “inflated” data:

1. We may think that our paid channels have a greater influence than the one they actually have.
2. We can end up assigning an excessively high volume of investment to certain channels because we are overvaluing their true contribution.

This poll was the igniter for this post, and as I start developing the topic we will see the variations that are generated when using one model or the other. After all, pears and apples are not the same and neither are last click and last non-direct click.

Let's go for it!

According to the information available in Google Analytics' support page, the last non-direct click model:

Ignores direct traffic and attributes 100% of the conversion value to the last channel the user clicked on before buying or converting. Analytics uses this model by default when it attributes the value of conversions in reports that are outside the "Multi-Channel Funnel" block.

Let's disect that.

What is considered direct traffic?

Traffic is classified as direct when a user searches the URL of your web page in the search bar without passing before through any channel. It is a form of non-paid traffic and in this case, the key point is that when landing on your website the parameter document.Referrer has no value or if it does, there is no search term associated to it. You have more details here about the entire flow that is followed in Google Analytics to categorize traffic.

What are multi-channel funnel reports?

These are reports included in the conversions section within Google Analytics, these reports allow you to compare different attribution models. You can choose out of a list of predefined options or create your own custom model adding all sort of interactions like impressions, clicks and direct traffic into the model.

One of the important pieces of the MCF models section is that you can adjust the lookback window in the report, ranging from 1 to 90 days.

What role does the lookback window play in an attribution modeling?

The lookback window can be defined as the period of time during which an interaction can receive credit for a conversion.

For example, in a 90-day window, transactions generated today are attributed to clicks or impressions that occurred up to 3 months ago.

To understand the importance of the lookback window for your business you can compare the revenue per channel with the last non-direct click model within MCF reports to the same model but analyzed in the reports of the other Google Analytics sections (for example, the acquisition reports or the custom ones), as a brief spoiler, even though it is the same model, you will find some interesting discrepancies.

The reason for that is that the reports in the multi-channel funnels section have a lookback window of maximum 90 days, while the rest of Google Analytics reports have a default lookback window of 6 months.

Yes, 6 crazy long months.

From what has been discussed so far there are two key points to keep in mind:

1. The default lookback window in the most common reports in GA is excessively long: up to 6 months.
2. All conversions generated through direct and in which there is prior interaction with other channels within a 6-month period are not attributed to direct, but to the previous channel.

Let's see it with an example: A user enters your website through a SEM ad and does not buy any product. After 6 months that same user returns to your website via direct and buys one of your products. In this path, according to the default model in Google Analytics, all credit must be assigned to SEM. Curious, right?

But imagine that at some point along those 6 months you chose to activate your retargeting campaign for this user who despite showing interest, chose not to buy. You keep insisting with your retargeting but the user does not click on your ad, until one day he chooses to buy your product and starts typing the name of your brand in his browser bar.

Once he lands in your site he buys the product he was looking for, but since the last channel he came through was direct and apart from that, last non-direct click does not take impressions into account, that sale will be attributed to SEM, even though there are at least two additional interactions of other channels after the paid search one.

One thing that's irrefutable is that if someone remembers or has saved the URL of your website, it is because at some point that someone has visited your page, either coming from another channel or because through other type of actions you have managed to generate brand recognition, but that certainly is not enough to consider that 100% of the credit of a sale must be assigned to the previous channel during a period of 6 long months.

Last click or last non-direct click, what is the effect in economic terms?

By analyzing the different models you will realize that the details literally make the whole difference. These are the results that we found when analyzing a GA account:

1. An additional 3% of transactions and revenue assigned to paid channels by expanding the attribution window from 3 to 6 months. Although the percentage is low, depending on the volume of your business we can be talking about just a few or a good couple thousand euros.
2. Between an additional 30%-40 % of transactions and revenue assigned to paid channels when comparing last click with last non-direct click.

I believe the numbers speak for themselves, but just let me point out how different your decisions could be if all of a sudden you took away 30% or 40% of the revenue from a channel.

I encourage you to do this exercise with your own GA data so that you see how important both the lookback window and the exclusion of direct as a channel are in the allocation of transactions and revenue for your business.

To do this, follow these steps:

1. Go to the Acquisition> All Traffic> Channels section and choose a time period, for example the last month.
2. Now, in another tab, go to the section Conversions> Multi-channel funnels> Model comparison tool.

• Select the same time range as in the acquisition report
• In the first selector choose the type of conversion that you analyze in your reports. In the case of an eCommerce it would be the transactions.
• In the second selector on interaction type, keep all interactions. In this way you are including the interactions of payment and non-payment in the report.
• In the attribution window select the maximum (90 days).

When choosing the models, select both last non-direct click and last interaction (if you want to analyze pure last click specifically, you will have to create a custom model).

Once you have them, download both reports. The Multi-channel funnel one can only be downloaded without sampling and usually it takes several hours to be available.

Once downloaded, you just need to calculate the percentage change of transactions or revenue based on the chosen model.

You can do it channel by channel or categorize the channels by paid / non-paid and make the calculation based on this grouping.

From my perspective there are two variations that are worthy of analysis.

1. What percentage of additional transactions or revenue get attributed when taking the lookback window from 3 to 6 months.
2. What percentage of additional transactions or revenue get attributed when excluding direct traffic from the model.

To calculate additional attribution for expanding the window from 3 to 6 months:

To calculate what percentage of sales has been over-allocated to paid channels due to the exclusion of direct in the conversion path:

In this last case you could use the data from the MCF report in both the numerator and the denominator. But most advertisers mainly make use of the reports that are outside of the MCF block, hence it will be way more interesting to calculate the variation with those reports instead of the MCF ones.

Why should you keep this information in mind?

Attribution models are a very useful source of optimization insights in the day to day of digital marketing professionals. Its main features are:

1. Help us understand if our actions in digital marketing are having an effect on generating revenue for the company
2. Facilitate the optimization of marketing investment, by giving us visibility on what works and what does not. In this way we can decide whether to increase, decrease or directly cut the investment in a certain channel or specific action.
3. Establish the final amount to be paid to the agencies and providers that have participated in the activation of digital marketing.

Point 3 is the one with the greatest impact. It is common for companies to outsource or seek additional support from agencies or third-party companies that charge a percentage of the revenue generated for the company.

It is also very common to establish the objectives of these agencies on the revenue generated by the channel they manage. The problem with that is that as we have just seen, the revenue generated by a channel can be up to 40% higher depending on the attribution model you use.

This is the main reason why, it is essential to ask ourselves what model we want to use to evaluate if our agencies, partners and providers are achieving the goals we've set in this type of agreements and also that we properly understand the implications of using one model or the other.

The post summarized in 3 ideas

1. Details make the difference: concepts such as the lookback window or whether or not the attribution model we use has the "non-direct" tag in its name can have a major impact on the data we analyze.
2. The last non-direct click model attributes up to 40% of additional revenue to paid channels with respect to the allocation made by the last interaction model or the last click model.
3. There is no perfect attribution model, although you can definitely find the one that better suits each situation. If possible, consider whether you really want to pay agencies and suppliers based on a model that assigns direct sales to activity generated 6 months ago.

What are your thoughts on this? Have you ever came across the difference between these two models? Did you calculate the effect this has on your business?