How to estimate the additional value of multi-channel attribution

In this article, we introduce the method for estimating the additional revenue you can gain with attribution models that consider the mutual impact of advertising channels. The article will be useful to analysts and marketing specialists who want to assess the potential benefits of moving from Last Non-Direct Click attribution to multi-channel attribution models.

In this article, we’ll tell you:

Problem Statement:

"What does that give us?" Fair enough, the question is asked by even those who are familiar with the shortcomings of standard attribution models. If you don’t know what attribution model is used in your project, it’s most likely Last Non-Direct Click—all the transaction value is attributed to the last channel that the user clicked through from before the conversion. This is the default attribution model in most analytical services, including Google Analytics.

It is a well-known fact that Last Non-Direct Click does not consider the mutual impact across channels. Do you know how to measure it in monetary terms? What share of the income does Last Non-Direct Click distort? How can we fix that?

Not all paths are equally useful

If you’re using Google Analytics 360 or OWOX BI Pipeline you have access to the unsampled data in Google BigQuery and you can test any hypotheses with comprehensive data. But if you’re using the standard version of Google Analytics, things can get pretty crazy finding the answer may not be easy.

Let’s take a look at the Path Length report:

Select transactions as the only type of conversion. As a result, you can see what share of income is generated by the orders made at the first visit. In our example (a multi-channel retailer with a moderate number of website visitors), multi-channel paths with two and more visits bring 72% of income.

But a 72% distortion seems unreasonably high; it’s hard to believe that even the Last Click attribution distorts the distribution of income so much.

Let’s take a closer look at what paths with two and more visits are the most popular. To do this, let’s open the Top Conversion Paths report:

Please note that:

  • The paths number 1,3,4,6,8 and 9 end with a Direct visit. The value in these sequences is assigned the same way as in Last Non-Direct Click model;
  • In the path number 7, the first traffic channel is organic search, the traffic channel that is hard to manage. Even knowing that its value is underestimated, we are unlikely to apply this knowledge.

This leads us to some important conclusions. To assess the potential of applying multi-channel attribution models, it is necessary to consider the sequences that meet the following requirements:

  • The value in the sequence must be assigned in a way different from Last Non-Direct Click attribution;
  • The manageable traffic channel must not be the last one in the chain.

Manageable channels include, at least, paid traffic sources. To highlight such sequences, we filter the channel sequences:

The following line should be set as the regular expression:

(Paid Search|Display|Affiliates|unavailable|Email|Other).*>.*(Paid|Display|Affiliates|unavailable|Email|Social|Organic|Referral)

In English, it means that the first source in the sequence must be manageable, followed by any source other than Direct. The matches between channel groupings and sources are provided in detail in Google Analytics Help.

After applying the filter, we get only those sequences in which the value can be assigned to manageable channels in a way other than in Last Non-Direct Click attribution.

Please note:

  • In default channel grouping, all the unrecognized traffic with utm-tags is considered unavailable, so we also take it into consideration;
  • If you decide to change the regular expression, do not forget that Google Analytics supports syntax RE2 only. It does not support certain expressions, such as exceptions (?!re).

Now let’s look at the Conversion Value in the Top Conversion Paths report. You will see that in this case, considering the mutual influence of of channels, you should redistribute $19,7M, which is 40% of the revenue:

Budget reallocation considering the mutual impact of channels, in our experience, results in 20% to 50% growth of ROAS in this segment.

As a result, the added value from the reallocation of your budget can be estimated as

Additional income = Redistributed income * ROAS growth

Redistributed income is the income that does not consider multi-channel impact when assessing channels with Last Non-Direct Click model. Such income can be reallocated.

ROAS growth is the efficiency gain in the given segment of income.

In our example the additional income is:
$19.7M ∗ {20%; 50%} = $3.8M to $9.6M per month.

To understand the impact such reallocation has on your your project, insert your own figures in the formula. Ask yourself, how would you feel if your sales decreased by this amount with the same advertising budget? What if the sales increased by the same amount?


Before applying the obtained results, it’s worth considering the impact of the following assumptions.

  • In Google Analytics Multi-Channel Funnels report, the paths are analyzed within a single cookie (clientId). Therefore, if your model uses combinations of user activities on multiple devices with userID (for example, in Google BigQuery), the average path length increases and the potential added value becomes even higher;
  • If you know in advance that investing more money in certain channels, e.g. brand channels or niche platforms, won’t drive any more traffic to your website, it is better to treat such channels as those that are hard to manage. It’s probably not reasonable to increase the budget for such channels;
  • Most importantly, if you have received a sampling message after applying the filter to the channel paths—congratulations, a bonus track is waiting for you!

1,000,000 is a fairly large sample. But since it is used to build a large number of small segments, there may be a significant final error. The unsampled results may differ by several times. To get the unsampled results, you need to eliminate segmentation for various paths from the report. Fortunately, this can be done with the help of Multi-Channel Funnels API even in the standard version of Google Analytics.

To do this, let’s create a report with the following parameters:

To see how channel grouping affects the outcome, you can add the following as the Dimensions:

Here’s the comparison of results with sampling and without sampling for a project with about 500K sessions per day:

Use of the findings

We have answered the question "What does that give us"? The next question is—"What do we do with it?".

First of all, congratulate yourself—your project has potential to grow, and there are more effective ways to invest in advertising.

The next thing to do is to determine the attribution model that you will use to evaluate multi-channel paths. We recommend the Funnel-Based Attribution method that attributes conversion value to your advertising campaigns according to their impact on users’ behaviour at each step of the sales funnel. In contrast to the associated conversions or applying several attribution models at the same time, this method does not attribute more than you have earned, ensuring full transparency of the calculation.

We will go over the details of the implementation and estimation of the method in one of the future articles. The main thing is to remember the general rule for assessing every multi-channel attribution model: you need to assess the changes on the website as a whole, not in a particular channel. Like in any team game, the goal of any business is maximizing the result of the team play. An individual achievement of a particular player—or a particular campaign, in our case—might not bring the expected results.

Used tools