The Challenge
Our client, a subscription-based retail merchant, noticed a gradual erosion in monthly active customers. Because user action logs, support calls, and transaction databases were siloed, the marketing desk could not isolate churn factors or identify at-risk buyers until cancellations occurred, leading to high cost per acquisition numbers.
The Approach
Predectivera unified these disparate pipelines into a cohesive database store. We built a random forest categorization algorithm to evaluate customer behavioral sequences and compute an individual "Churn Score" from 0 to 100. This logic was connected directly to interactive Power BI dashboards, making at-risk accounts instantly visible.
Implementation
We wrote SQL parsing views to combine data from Postgres ledger records, Shopify client APIs, and Zendesk ticket databases. The prediction logic, compiled in Python with pandas and scikit-learn, runs nightly inside a Docker task on AWS. Clean cohorts sync straight into a Microsoft SQL server which updates the dashboard client interface every morning.
The Result
Our dashboard gave account managers the details they needed to launch automated customer recovery emails. Within 90 days of launch, client cancellations decreased by 32%, securing significant recurring revenue. Model predictions maintain a 91.4% accuracy threshold on weekly cohorts.
"Predectivera transformed our data mess into a structured, predictive dashboard engine. We went from guessing monthly renewals to predicting customer churn with over 90% accuracy."
Sarah Jenkins VP of Customer Success, Retail Plus