Customer Data Analytics Overview
SessionM strives to make data actionable in real time, with the goal of allowing marketers to create engagements powered by data, that facilitate personalization at an individual customer level. Using purchase and activity history as well as other latent factors, the SessionM Platform is able to predict customer preferences that influence affinity, motivation, and intent to purchase.
Leveraging this data, the platform is able to provide a number of metrics, specifically:
RFM Metrics – Insight into customer purchase recency, frequency, and monetary spend.
Customer Lifetime Value – Forward looking metric that projects revenue that each customer will generate over their lifetime with a brand.
Probability for Churn – Likeliness that a customer will or will not return to a brand.
Product Recommendation Scoring – Identified product affinities on both individual customer and audience levels.
RFM is an acronym for Recency, Frequency, and Monetary Spend. These historical metrics are common for many marketers, and are defined as:
- Recency – The time since a customer performed a specific action (e.g., purchase).
- Frequency – The number of times an action was performed.
- Monetary Spend – The dollar amount associated with the customer action.
The SessionM Platform calculates RFM for three distinct time periods: 7 day, 30 day, and lifetime. Using RFM, it is easy to compare the historical performance of one customer to the full customer base.
RFM Within Customer Profiles
RFM metrics can be viewed for customer profiles within the Performance Metrics section of the Customer Profiles Module, providing a view of where each individual customer stands in comparison to the rest of the customers in your program.
We can see through these metrics that Andee Solddonna has performed 2 transactions totaling $37 over her known customer lifetime, with the last transaction taking place 6 weeks ago. This data can be matched with others to reveal that Andee has spent more and purchased more often; however, the date she performed her last action is not as recent as the median for the customer base.
RFM Within Audiences
The SessionM Platform also allows you to leverage RFM metrics within the Audiences Module to define an audience with various recency, frequency, and monetary spend attributes during creation.
As seen in the above example, a custom audience can be built using these metrics to target those who have made a purchase within the past 21 days, while also making over 5 purchases and spending more than $100 over the course of their lifetime with the brand.
These metrics are intended to review historical customer activity, as RFM does not predict future value or transaction flows. In order to understand future value, another metric is needed, specifically Customer Lifetime Value.
Customer Lifetime Value (CLV)
Customer lifetime value is a forward looking metric that calculates the projected revenue each customer will generate over the course of their lifetime with a brand. The metric can also be seen as the present value of future cash flows attributed to the customer relationship.
SessionM’s CLV Approach
The basis behind the SessionM Platform’s machine learning model is that every customer can choose to purchase/interact with a brand at any point in time. Essentially, each customer has two distinct coins. The first coin determines if the customer recalls the brand (or not), and the second coin determines if the customer buys (or not). The probabilities associated with these coins are unique to each individual. Using historical data along with these models, the platform is able to predict:
Expected Future Transactions – How often each customer will interact with the brand.
Customer Lifetime Value – How much each customer will spend when they transact with the brand.
Churn Probability – Which customers are likely to return in the future.
These metrics can be visually defined through the image below:
In this example, each star represents an equal-value historical purchase for both Customer A and Customer B, with the vertical line representing the present day. Both customers have the same recency, as the timing of their most recent purchase is the same. Looking at each customer solely through the lens of RFM, Customer A has a higher value to the brand because he/she has made more purchases (e.g., higher frequency) and spent more money (e.g., higher monetary spend).
However, when looking at the value of these customers through the lens of CLV, we come up with a different conclusion. Customer A has a higher propensity to churn because his/her “normal” transaction behavior has been broken resulting in a less likely chance that Customer A will return than Customer B. Additionally, since Customer A’s churn value is high, his/her CLV score is reduced, leaving Customer B with a higher projection for customer lifetime value moving forward.
As the platform’s model for these metrics is statistically based, it is easy to understand and measure the effect of any behavioral or mathematical changes to it and then make the necessary adjustments.
CLV Within Customer Profiles
Once calculated, marketers can view an individual customer’s value within the Performance Metrics Section of the Customer Profiles Module.
Along with a CLV value metric, lifetime value is also surfaced within the platform as a rank (Low, Medium, High) to provide a comparison of the customer’s value to the rest of the customer base.
Probability for Churn
As mentioned in the CLV overview, part of the SessionM Platform’s process to determine customer lifetime value is establishing the likelihood of a customer to return. Quite often, enterprise brands are in non-contractual settings where customers can come and go as they please and there is no “churn event” to notice. Instead, we calculate the probability that an individual customer has churned.
SessionM’s Churn Model
The platform’s churn model looks at the historical transactions at an individual customer level to understand the “normal” behaviors and cadence for each customer. As long as that “normal” pattern is maintained, churn probability will remain low. Once the model senses a negative deviation in that “normal” pattern, the churn probability will increase.
As an example, if a customer shops at a store once a week, the model will maintain low churn scores for that customer until a full week passes since their last transaction. At that point, the churn probability will increase rapidly for each day that passes. Similarly, if the customer increases frequency from once a week to twice a week, the model will notice and adjust the “normal” pattern to fit this new cadence. As more and more data is fed into the model, it becomes more accurate in determining what is “normal” and what is “abnormal,” a judgment that can justify sending relevant alerts to marketers.
Churn Within Customer Profiles
Much like the other customer analytics provided by the SessionM Platform, probability for churn is surfaced within the Performance Metrics section of the Customer Profiles Module.
The metric is displayed as “Risk of Churn” with statuses of “Low,” “Medium” and “High” – all of which adjust based on the pattern of observed customer transactions.
SessionM has a robust recommendation engine that is able to provide scoring at scale for large data sets. The platform’s engine uses machine learning to deconstruct historical purchase data into latent factors that can inform estimates of previously unknown customer/product affinities.
SessionM’s Recommendation Engine
The recommendation engine uses a matrix factorization with an implicit feedback model similar to what is used in various industry-leading media recommendation tools. This model does not require extensive knowledge on either customer demographics or product attributes in order to provide significant results, and is able to generate personalized recommendation scores on a per-customer and per-product basis as the output.
Product Recommendations Within Customer Profiles
Scores generated from the recommendation engine, along with associated probabilities, are leveraged to provide product recommendations at an individual customer level through the Customer Profiles Module.
Through these recommendations, important marketing questions such as “Which products is this customer likely to enjoy with a confidence score over 80%?” are able to be answered. Additionally, by analyzing product recommendation correlations, the platform is able to provide item-pairing recommendations which can be used for various cross-sell or upsell promotions.