Multi-Touch Attribution Model - Identifying the right channels

In the ever-broadening media landscape, succeeding on a large scale rarely depends on a single channel: a combination of marketing activities generates awareness and motivates consumers to direct their choice towards a particular brand. Each of these activities may have an impact of a certain extent and generate more or fewer conversions, i.e. product purchases.

Identifying the right channels with Multi-Touch Attribution Models to reach consumers is becoming increasingly important to improve the efficiency of the marketing spend. Knowing where the consumers came from when they made a purchase and which of the brand’s paid or owned media they had seen beforehand is called Media Attribution.

Table of content:

• Why use Attribution Models in Media?

• Why does it make sense to try to improve the conversion path?

• Why is Machine Learning needed for Multi-Touch Attribution?

• How is Multi-Touch Attribution affected by the death of Cookie-ID?

• What are the alternatives to Cookie-IDs in Multi-Touch Attribution Models?

Why use Attribution Models in Media?

Conversions are rarely determined by the last touchpoint: consumers tend to have multiple encounters with a brand before making the decision, which is usually also correlated to the investment that the product represents to them. Therefore it would make little sense to focus the marketing budget only on one unique media.

There is a need to estimate the impact of each medium on the overall advertising ROI, and when it comes to digital channels, they also offer the opportunity to deep dive into the data with more granularity. Due to the fragmentation of digital communication routes, companies are increasingly struggling to be visible to consumers without heavily investing in cross-device and omnichannel advertising campaigns, which can entail investments that need to be kept under close control.

The available technology makes it possible to trace an individual conversion path within all digital environments and follow the steps made by the consumer over an “extended” period of time. Brands can then identify the sequence and the channels which have worked best to influence the progress towards the purchase. 

This approach helps companies optimize their marketing spend and target the right customers in the right places and at the right time.

This is called the Multi-Touch Attribution Model (MTA), and it is as complicated to really achieve as the name is difficult to pronounce. Let us explain why.

Why does it make sense to try to improve the conversion path?

Because every nudge in the right direction - as long as it happens at the right moment - increases the chances of conversion. In the extremely competitive landscape of the modern digital space, brands constantly fight to retain their customers.

Any incentive that can contribute to gaining new customers helps maintain the brand in a strong position.

It is extremely difficult to anticipate the conversion rate on a new campaign, one can compare with competitors or base oneself on industry benchmarks; but after all, it is the action that would provide your truth, from there, improvements on your conversion rate will be your key to success.

Why is Machine Learning needed for Multi-Touch Attribution?

Imagine a labyrinth of never-ending trails, deceptive turns, and dead-end routes… This is what a consumer’s purchase journey looks like in our current digital age: a maze where the consumer bounces between different websites, social media, and devices before he or she decides whether to purchase the product or not.

• Data is collected across devices based on cookies: it has to be sorted and organized in order to be sequenced.
• Not all media activities have the same weight: seeing a display ad and spending 3 minutes on a website clearly have a different impact on the reminiscence of a brand.
• Memories diminish with time at a non-linear pace, it is offset by the multiplication of exposures, creating a build-up effect that needs to be taken into account.
• The number of possible touchpoints and formats is great.

It means zillions of potential routes to conversion that need to be sequenced, weighted, and analyzed to create a heat map of the different opportunities for brands to improve their current strategy.

How is Multi-Touch Attribution affected by the death of cookie ID?

The way Multi-Touch Attribution Models are built today relies heavily on Cookie-IDs. Their disappearance means the death of the Multi-Touch Attribution Model as it is now: a cookieless world would influence the granularity and precision of information collected as well as the depth of the path that may help the model to predict the weight of different media channels.

What are the alternatives to Cookie-IDs in Multi-Touch Attribution Models?

More First-Party Data Collection

Nowadays most marketers have already given up the usage of third-party data (cookies) since it’s slowly dying out anyway and switched to first-party data.

First-party data is more accurate and reliable than third-party data, allowing advertisers to create a better-tailored customer experience.

By utilizing all of the existing touchpoints you have with your consumers and prospects, you will be able to aggregate the data on a customer level in order to draw the path that leads to purchase at the end of the funnel, which is the key to MMTs Multi-Touch Attribution Model to predict different dimension based on these paths. We will cover the first-party data strategy in an article soon.

Universal ID

A lot of frameworks use universal identifiers to encourage the creation of a unique user ID for each online user. The solution helps preserve the marketers' ability to deliver targeted advertising while giving users more control.

When a user visits a website or app that supports universal identifiers, the single sign-on technology helps to capture the user’s email address along with the one-time consent.

Probabilistic modeling

Probabilistic modeling links a single user’s engagements across numerous devices to a unified customer profile by utilizing prediction algorithms to correlate information such as IP address, operating system, location, wifi network, and behavioral data to a profile at a specified confidence level.

Using this technique, you will be able to tie all the information together such as touchpoints of a specific user, and end up with a full path that can be later used in MTA modeling.

Fingerprinting

Fingerprinting is the process of creating a digital fingerprint of a user that can be traced across channels by employing a cross-section of non-personally identifiable information, such as device type and location.

Just like probabilistic modeling, this approach enables analysts to extract the necessary information like touchpoints to feed it into the MTA model.

How can MMT help you?

MMT combines its media and data science expertise to develop its own Multi-Touch Attribution Models adapting to market evolution.

Our solution integrates two models addressing the sequencing, memory build-up and decaying, influence of the advertising formats, and binary conversion degree. It offers clear visualization of the most efficient paths as well as the contribution of each media to the conversion.

Our model requires the input of at least 3 to 6 months’ data on digital activities in order to generate the most accurate results. Our standard model can be run in a matter of days and visualized in a web-based dashboard to monitor the progress.

Need more details? Feel free to get in touch with us.

Sources

• https://medium.com/machine-learning-for-business-problem/how-to-implement-an-attention-rnn-into-solving-the-multi-channel-attribution-problem-6fa90d935859
• https://marketinginsidergroup.com/demand-generation/are-your-lead-conversion-rates-above-average/