Blog Contributor: Ron Sadi

We often hear from clients, that maximizing the full value of their customer data is challenging. Many brands only use a fraction of their first-party data to inform marketing programs, leaving real dollars on the table. What can enterprises do differently to leverage proprietary data to improve their marketing curriculums and increase customer lifetime value?

It begins and ends with gaining a deeper understanding of individual customers’ behaviors, interests and intent; and actioning off of those insights to personalize one-to-one marketing at scale. The following case study illustrates how Zeta leveraged Natural Language Process (NLP) and machine learning to unlock value from existing customer data for a leading pet retailer in the U.S.

 

Business Challenge: Staying Competitive When eCommerce Giants like Amazon Are Creating Downward Pricing Pressure on Traditional Brick-And-Mortar Retailers That Are Vertically-Oriented

Traditional retailers today are challenged with staying competitive against the convenience and downward pricing pressure of online distributors, like Amazon. This pet retailer differentiates itself by driving customers to products and services that are unique to their brand and only offered in-store, specifically: pet grooming and animal care services. Therefore, there was a strategic imperative to identify and target consumers who had a propensity for these in-store activities.

 

Proposed Solution: Leverage NLP and Machine Learning to Identify Qualified Customers

How did we identify customers who were in-market for pet care services? We leveraged NLP and machine learning to harvest behavioral data across the internet to identify clusters of individuals who had a propensity for pet grooming and animal care services.

There are three pillars to this solution:

  • Understand content consumption signals. Specifically, what content are users consuming online and what’s the frequency and recency of that consumption
  • Identify real individuals deterministically through a persistent identifier, typically an email, rather than a probabilistic approach that relies on non-persistent data, usually browser cookies
  • Connect (1) our understanding of consumers’ behaviors and interest to (2) real individuals to (3) generate actionable high-value audiences

Findings: Highly-Active Pet Owners had the Highest Propensity for Pet Grooming Services

By mapping the behaviors and interests of the retailer’s existing pet grooming purchasers, we found that they tend to be highly-active, outdoorsy, and hygiene focused individuals, as seen in the graph below. Those categories with higher interest levels indicate a more significant proportion of the pet retailer’s customers that we see on average:

Methodology:

Understanding Content Consumption Signals

To unlock the value of behavioral signals, we first need to monitor millions of user touchpoints generated as consumers move across the internet each day. Zeta sees over 600 million signals each day across 4.5 million websites. Listening for these signals in real-time, enables us to derive intent and proactively identify consumers as they move into their purchase cycle rather than react to passive ‘data-at-rest’ e.g., transaction history.

In the case of our pet retailer, we saw up to 20 million behavioral signals generated by 3 million of their customers on our network each day. Those signals included:

  • The type of content customers consumed
  • The subject matter of that content
  • The frequency and recency of that consumption

Leveraging NLP and Machine Learning to Decipher Signals

Some signals are loud, and some signals are latent. This is where Zeta’s proprietary NLP comes into play. We leverage NLP, AI and machine learning to decouple those behavioral signals and derive consumer intent, allowing us to ultimately generate a highly-targeted actionable audience of pet grooming intenders.

We do this by continuously clustering and classify relevant keywords and topics (pet nutrition, fur, animal health, etc.) to generate custom audience clusters and then score each individual based on their semantic relationship to those segments. In this case, we used “pet grooming” as our primary taxonomy which we then layered with customer profile traits.

To close the loop, by continually monitoring outcomes through machine learning, Zeta can interpret the value of loud and latent signals and perpetually optimize audiences.

Identifying Real Individuals

The challenge digital marketers face today, is knowing if their messaging is reaching real individuals. The majority of programmatic advertising relies on a probabilistic method of identifying users based on a perceived connection to a device or browser. These identifiers are non-persistent as users clear their cookies and move across devices, resulting in low-precision targeting that’s not truly 1-to-1.

On the other hand, a deterministic approach uses persistent identifiers that require a defined link, e.g. as users are logged in to their devices via email. This allows us to activate omni-channels marketing curriculums with high-precision enabling 1-to-1 at scale.

 

Summary: Signals, Identity and Connectivity Drive Revenue

When marketers examine these three fundamental principles: SIGNALS, IDENTITY, and CONNECTIVITY, they can identify high-value audience groups by leveraging NLP, AI and machine learning by (1) monitoring signals to understand consumer interests and intent, (2) map that understanding to a real individual and (3) generate highly-targeted audience groups relevant to the unique business objectives they are trying to solve. Using these principles, Zeta identified 3 million customers that were actively seeking pet grooming and animal care services and executed 1-to-1 marketing programs that incorporated unique creatives and personalized journeys based on the best messaging, best time and best channels.

%d bloggers like this: