Understanding Marketing Channel Attribution with Markov Chains

Author: Rahul Chaudhary

I recently watched a comprehensive YouTube video by Jessica Duncan, a marketing data scientist, focusing on the intricacies of marketing channel attribution using data science techniques, specifically leveraging R and Markov chains.

At its core, marketing attribution aims to determine which customer touchpoints play a pivotal role in converting leads into customers. These touchpoints can span various channels, such as social media, email marketing, or customer service interactions, and can even include untracked influences like word-of-mouth referrals.

Duncan provided a scenario involving a company selling baby shoes, illustrating the customer journey through touchpoints like an Instagram ad and a follow-up cart abandonment email. This two-touchpoint story highlights a common marketing challenge: understanding the interplay of channels rather than attributing success solely to the first or last touchpoint.

Current Attribution Models

Duncan discussed the limitations of traditional attribution models, primarily first-touch and last-touch attribution methods. The methodology assigns 100% of the conversion credit to either the initial or final touchpoint, resulting in a significant oversimplification. For instance:

• First Touch: If the customer discovered the product through an Instagram ad, all conversion credit would go to that ad, ignoring subsequent influences.

• Last Touch: Conversely, if the purchase followed an abandoned cart email, it would receive sole attribution, disregarding earlier touchpoints.

Both models ignore the complexity inherent in customer behavior. The reality is that a consumer could interact with multiple touchpoints before making a purchase, making a more nuanced model necessary.

Moving Beyond Simple Attribution

As Duncan pointed out, many marketing departments still rely on these simplistic models. A step up is multi-touch attribution, which attempts to distribute credit among various touchpoints. Yet, this is often still formulaic and lacks a deep understanding of customer intent.

Multi-touch models include:

1. Linear Attribution: Assigns equal credit to every touchpoint. This approach is excessively simplistic because it fails to account for varying levels of intent across touchpoints.

2. U-Shaped Attribution: Allocates higher credit to the first and last touchpoints, leaving a smaller portion for interactions in between.

3. Time Decay: Prioritizes recent interactions, giving more credit to touchpoints closer to the conversion event.

These models still lead to inaccuracies as they do not adjust for differences in customer journey lengths or specific layer dependencies. A parent scrolling through Instagram might only engage with the ad for a moment, whereas someone completely different may require several touchpoints spread over weeks or months.

The Markov Chain Approach

Duncan introduced Markov chains as a promising alternative that offers a more sophisticated understanding of customer journeys. Markov chains account for the probability of transitions between different states (or channels).

For example, consider three states of a customer: “showing interest” (smiling), “losing interest” (crying), and “purchasing” (asleep). Each probability needs to sum to one, meaning every customer either stays in the same state or transitions to another:

• Staying in state: 32% (smiling)
• Transitioning to crying: 26%
• Transitioning to asleep: 42%

For marketing attribution, the different channels are modeled this way, providing probabilities for transitions from one channel to another (e.g., from a social media ad to an email).

Implementing the Markov chain model allows marketers to better predict customer behavior and ascertain which channels most effectively lead to conversions.

Practical Implementation with R

Duncan walked through how to set up this analysis using R, particularly through a package named ChannelAttribution. Beginning with clean data, users can format their touchpoints, allowing professionals to estimate customer interactions effectively.

Using a simulated example with roughly 6,500 customers and 50,000 touchpoints, she illustrated that most customers interacted with between seven and eight touchpoints leading up to a conversion or abandonment.

The beauty of the Markov model lies in its ability to distill complex touchpoint interactions into actionable insights. With a probabilistic framework, marketers can answer critical questions:

• Which channels should receive increased budget allocation?
• Where can spend reductions occur without significant drops in conversion rates?

Furthermore, the transition matrix produced from Markov chain calculations can show movement probabilities across channels, offering a clear picture of where customers are likely to engage the most.

Key challenges include ensuring data availability, which can be complicated by privacy regulations, and accurately modeling customer journeys that contain necessary prerequisites (e.g., telephone consultations before a product sale in certain fields).

Wrap-up

The Markov chain model promotes a holistic view of the customer journey, intertwined with proper attribution strategies. By adopting such advanced methodologies, marketers equip themselves with insights critical to refining budget allocations and improving conversion rates in a data-driven landscape. This approach not only clarifies the effectiveness of each touchpoint but also sheds light on the interdependencies within the overall marketing strategy.