Engagement Metrics & Segmentation
Engagement Scoring offers a Data Science-driven customer segmentation service/product. We have developed engagement profiles of individual customers based on web (Olytics) and email (Email Builder) interactions. Profiles consist of various engagement metrics derived from amount and timing of interactions. Once engagement metrics have been generated for all relevant customers an unsupervised machine learning technique called clustering is implemented to segment the customers into broader levels of engagement. The results are recalculated and updated nightly and can be queried in Audience Builder. Engagement Scores are calculated separately for both web and email activity.
Engagement Metrics
We measure engagement based on five factors: total # of interactions (aka “volume”), frequency, recency, intensity, and momentum.
Volume measures the amount of positive interactions a customer has had with a client
Frequency measures the rate of positive interactions that occur for this customer over time
Recency measures how recent a customer has interacted with client
Intensity is analogous to “depth” in relation to engagement. It measures how deep a customer engages each session (session=day).
Momentum is a ratio of a customer’s recent interaction compared to that customer’s historical interaction. This metric prioritizes recent engagement compared to usual engagement.
Once values are calculated for each customer, engagement metrics are cleaned for extreme outliers (z-scores < 3) then normalized (using boxcox transformation) and finally binned into relative “scores” ranging from 0 – 100. 100 being the most frequent/recent/intense visitors.
Important note: a customer may have several different scores depending on their activity. For example, if John Doe visited a web page and opened an email within a Profile of Defined Customers, John will have 2 row of scores (1 for each web and email) representing their relative engagement.
Customer Segmentation Clustering
Within each channel and profile an unsupervised learning algorithm is used to group customers by their type of interaction with a client’s brand. In aggregate we have found 4 meaningful segments of customers based on engagement that can be seen below.
Cluster | Cluster Name | Description |
|---|---|---|
A | At – Risk | At-Risk customers have a measurable history of engagement, but little to no activity in last 2 weeks. |
B | Consistently Engaged | Consistently engaged refers to customers with higher total # interactions, frequency, and intensity scores. These customers are visiting often as well as consistently over a longer period of time. |
C | Recently Engaged | Recently engaged refers to customers with highest recency and momentum scores. These customers are the most recent visitors that have significantly increased engagement in the last 14 days. |
D | Unscored | Unscored customers have either never opened an email or have recorded less than 3 olytics behaviors. This is not enough activity to generate meaningful engagement scores. |
Overall Engagement Scores
An added feature to engagement scoring is the “overall engagement score” that is an average of a customer’s web and email scores (note: not all customers have both web and email activity so they will not have an overall score). This metric aims to measure a customer’s level of engagement across both web and email channels. The overall score records are then also put through the clustering algorithm to assign overall scores to the same 4 segments seen in web and email.
The overall record is then assigned to a cluster in the same fashion as web and email.
Audience Builder
Audience Builder will have the following fields available for querying within the “Data Science” folder:
“Web Engagement” Skittle which contains the following fields:
Web Frequency Score (numeric field, valid values 0-100) – should allow for numeric ranges
Web Recency Score (numeric field, valid values 0-100) – should allow for numeric ranges
Web Intensity Score (numeric field, valid values 0-100) – should allow for numeric ranges
Web Momentum Score (numeric field, valid values 0-100)- should allow for numeric ranges
Web Cluster (multi select from clusters available) – should show Cluster Names as selectable
“Email Engagement” Skittle which contains the following fields
Email Frequency Score (numeric field, valid values 0-100) – should allow for numeric ranges
Email Recency Score (numeric field, valid values 0-100) – should allow for numeric ranges
Email Intensity Score (numeric field, valid values 0-100) – should allow for numeric ranges
Email Momentum Score (numeric field, valid values 0-100)- should allow for numeric ranges
Email Cluster (multi select from clusters available) – should show Cluster Names as selectable
“Overall Engagement” Skittle which contains the following fields
Frequency Score (numeric field, valid values 0-100) – should allow for numeric ranges
Recency Score (numeric field, valid values 0-100) – should allow for numeric ranges
Intensity Score (numeric field, valid values 0-100) – should allow for numeric ranges
Momentum Score (numeric field, valid values 0-100)- should allow for numeric ranges
Table of Contents