Introduction

One of the biggest challenges in digital marketing is determining the contribution of each of your marketing efforts to a successful campaign. Understanding the value of each marketing channel allows you to focus your efforts on those channels that generate the best outcomes and target your budgets more effectively.

Background

’Channels’ are the various routes through which traffic arrives at your website. The default channels in Google Analytics are Paid Search, Social, Organic Search, Direct, and Other. Their companions, ‘Source’ and ‘Medium’ describe more specifically the origin of your traffic, and the means by which it arrived at your site. Examples of ‘Sources’ are Twitter, Google, and MailChimp (i.e. platforms on which you are visible to your customers) and examples of ‘Mediums’ are Organic Search, Paid Ads, and Email (i.e. the publishing method through which the ‘Source’ made you visible to your customers). ‘Mediums’ are like the default ‘Channels’ mentioned above but are slightly more descriptive and specific.

Very often, we’re interested in reporting the number of visitors, e-commerce transactions, or goals that have come from each channel. For example, if you’re investing heavily in Google Ads advertising you want to know that you have a positive return on your investment and be confident that it’s driving the outcomes you want to achieve compared to other, free marketing channels such as social media posts.

Google Analytics

Google Analytics reports on the number of sessions, page views etc. that derived from each channel. However, this assumes that each channel was the only one that the user interacted with on their journey to your site. Actually, we know that users can be exposed to a variety of marketing channels in their journey, each of which might nudge them a little further along the path to visiting your site, and ultimately to them making a transaction or achieving a goal. If you’re considering only the first or last marketing channel the user followed, you’re ignoring the true contribution of all the other ‘touchpoints’ and might be underestimating their real value.

Google Analytics does have a multi-channel funnels report that can help you understand the different paths your users took to reach your site and meet your goals. This report also provides some deeper insight into the importance of each stage of the path. In addition, the attribution report allows you to weigh the value of each channel based on whether it was the first or last touchpoint in the user journey, or if it lay somewhere along the line.

Added Sophistication – Markov Chain Modelling

Google Analytics is great for a quick overview of your channel attribution.  However, there are alternative methods available that provide greater flexibility, more insight, and greater accuracy than is currently provided by Google.

At HMA we employ a statistical method known as Markov Chain modelling to better understand the contribution of different marketing channels to the path users follow to reach goals. With enough visitor behaviour data (extracted automatically from Google Analytics), Markov Chain modelling assigns probabilities to the movement of users from one stage in the path to another and takes a non-biased approach to calculating the contribution of each marketing channel in the path. As your site receives more visits and conversions, the model becomes more accurate.

An example:

In the example below we’ve chosen to measure the goal of visits to a specific web page using ‘AdContentPath’ – a built-in dimension in Google Analytics with identifying information on the different emails and social media platforms used to drive conversions.

With detailed analysis, we can calculate the relative value of each channel and social media post (source) we used to drive traffic to this web page.

In the table below, the results of our Markov Chain modelling are show in in green.

• Column A shows the source or medium of goal conversions
• Column B shows our calculated number of goal conversions attributed to each entry in column A using Markov Chain Modelling
• Column C shows the figure in column B as a percentage of all goal conversions
• Column D shows the number of goal conversions where the source/medium was the final touchpoint in the user journey, ignoring those that came before
• Column E shows the number of goal conversions where the source was the first touchpoint in the user journey, ignoring those that came after
• Column F shows a calculation of the average number of goal conversions where the source contributed at any stage in the user journey

We can see that ‘Email 2 List A’ delivered the most (almost a quarter of) goal conversions. If the goal in question had a value associated with it (e.g. the sale of a product) this data could also be included in the model, and an associated ‘Markov Value’ calculated for each source/medium which would support return-on-investment (ROI) decisions

As you can see, the conversion rates for each of the sources broadly agrees across different models. However, notice that in some cases (blue boxes) the first- and last-touch conversions disagree; under normal circumstances how would you decide which is the best model to pay attention to? This is where the linear touch model usually comes in as this considers the value of channels appearing at some point in the path with an equal weighting given regardless of what stage it appeared.

So why not just use Linear Touch Conversions?

Whilst our analysis includes the familiar ‘first touch’, ‘last touch’, and ‘linear touch’ measurements available in Google Analytics, these differ from our Markov conversions as we take a more sophisticated approach to weighing the value of each component of our path. This provides a far more refined measurement than the Linear touch model, based on the way it calculates the contribution of each source to different stages of the user-journey. Even small improvements can have a big influence on success rates and budget allocation decisions so it’s important to get as granular as possible.

You might also notice that our Markov modelling has assigned value to some of the lower-ranked sources, whereas the other models have not. Although these contributions are small, they demonstrate an example of ads that did in fact contribute to the goal conversion at some point– an important observation that was missed by the standard Google Analytics models.  Whether these small contributions were worth the cost of the ‘AdContentPath’ is an ROI decision that can then be considered.

In our example, we’ve not only gained deeper insight into the contribution of each marketing channel, we can also go back and look at the individual social media posts and emails to figure out which messages and what language worked most effectively. We could, for instance, choose to review the length, language, and call-to-action of the most successful posts and emails, and apply the same principals to future campaigns.

Summary

In summary, Markov Chain modelling provides a more sophisticated assessment of the performance of your marketing channels in driving users to a desirable endpoint. This helps you assess the true value of each channel and allows you to reassign budgets and campaign activities in a more efficient way.

Bear in mind that scheduled campaigns over a set period (e.g. a short, targeted Facebook campaign) will influence the long-term output of the model, so the timing and success of these types of campaigns must be considered when reviewing the results. Over the short-term, it’s immediately obvious if these campaigns contributed to driving more conversions.

For more information on how we can add insight to your online business goals, get in touch here.

Glossary

Channels – the routes through which you are visible to, and communicate with potential and existing customers

In Google Analytics these are Paid Search (advertising), Social (Twitter, Facebook etc.), Organic Search (Google, Bing, Yahoo etc.), Direct (users coming straight to your website), and Other

Source – the origin of your web traffic; the specific platform from which it originated

e.g. Google, LinkedIn, MailChimp

Medium – how the traffic was delivered via the Source

e.g. organic search, a paid advert, referral from a link in a social media post or email

Multi-Channel Funnel – a feature of Google Analytics that allows you to visualize and quantify the number and type of steps users took to reach your website or convert on a goal

Attribution Model – a computational method for analyzing the contribution and value of something to a chosen goal or action, e.g. “how much has Facebook contributed to diving visitors to make a purchase on my website, compared to Google?”

Markov Chain Modeling – a mathematical technique that calculates the probability of moving from one state to another along paths of different lengths, based on a collection of data from different paths. In the context of a user-journey, this means how users responded to seeing different web content relevant to your site, delivered through different channels, and how effective each was in driving that user to a desired outcome, e.g. visiting your site or purchasing an item from your web-store.

Dimensions – groups from Google Analytics that define categories of data, e.g. Source, Medium, AdContentPath

Last-touch conversions – a measurement of the contribution of each source, medium etc. that was the final touchpoint in the user journey, ignoring those that came before

First-touch conversions – a measurement of the contribution of each source, medium etc. that was the first touchpoint in the user journey, ignoring those that came after

Linear conversions – a measurement of the contribution of each source, medium etc. that appeared at any stage in the user journey, given an equal weighting regardless of what stage it appeared

Return-on-investment (ROI) – the net value outcome of an activity, given the associated cost of developing and deploying it