Client
Our client is one of the leading Martech agencies in the USA, specializing in promoting subscription-based businesses. Their clientele across the US publishing industry constitutes 60% market share. The company delivers a full cycle of marketing, business development, and business automation services, ranging from strategic consulting to building long-term marketing and selling strategies, implementation and optimization of tech stack, and more.
Challenge
The client used several off-the-shelf tools for audience analytics, but those tools provided a restricted set of metrics. Lightpoint team was tasked to develop a custom system that would calculate and predict additional metrics, churn rate in particular, to enrich customer analytics and empower data-driven decisions.
Project Description
It was agreed to develop a server-side solution that would communicate with the website through an API and web plugin. It is the server where data is collected, analyzed, and stored, as well as where a predictive model operates. On the other end, the web plugin monitors website visitors and calls an API to retrieve historical data for each visitor (whether one is authorized or not) from the server. If any, the web plugin places records in the data layer, from where the client employees can either retrieve this data for decision-making or redirect it to customer engagement software and traffic analytics tools (Google Analytics and Adobe Analytics are among them).
To prepare algorithms for data collection, analysis, and predictive model training, our data scientist analyzed and identified a set of metrics that correlate with churn. Based on that, two types of data were defined for analysis:
Unauthorized, anonymous website visitors activity
is used to study user behavior on pre-subscription phase – in other words, to collect some pre-history of subsequently authorized subscriber behavior. The system collects such events as page views, session duration, visit frequency, clicks, and more.
Authorized website users (subscribers) activity
is used to calculate and predict churn rate and churn percentile in real time. Among major items that the system monitors are subscription details, duration, payment regularity, payment history, purchase history, mismatches between user location and local subscriptions, and more.
Once the data is collected, it is handed over to the ML model to study and refine predictive algorithms for maximum accuracy.
Apart from churn rate, the system calculated and predicted churn percentile. Compared to churn rate, which measures the proportion of customers lost over a specific period, churn percentile refers to the percentage of customers who have stopped using a service or product, ranked against others in terms of the likelihood of churning. This is particularly useful for comparing churn behavior across diverse groups, allowing for targeted interventions based on relative churn risk rather than absolute numbers.
Key Features
Churn rate and churn percentile calculation per each user
ML-driven prediction of churn rate and churn percentile
Anonymous website visitors identification and tracking
Integration with Google Analytics, Adobe analytics, other traffic analytics tools, as well as customer engagement software
Team & Time Frame
The team worked on this project for 6 months, and included:
1 Data Engineer
1 Data Scientist
1 Full-stack Developer
1 Project Manager
Major Tech Stack
Type Script, Google Cloud, Google Data studio, Python, Go (REST WebAPI), NodeJs, Google bigquery, Mongo, Postgres
Results
The system we built enabled our client to track churn rate and churn percentile in real time to make informed decisions for further customer retention strategies and get a bigger picture of audience behavior (which off-the-shelf analytics solutions weren’t able to embrace) to improve performance and accuracy of all other marketing and sales strategies.
