The workflow before
TechSmith's usage data from Snagit and Camtasia, e-commerce transactions, support tickets and marketing data lived in separate systems. Analysts stitched them together by hand, spending 30+ hours a week on reports, and dashboards ran on daily batches. There was no view of a customer from trial to renewal and no way to spot accounts likely to churn.
What we built
We built a data lake on Azure Data Lake Storage with Databricks on top, using a bronze, silver and gold layout. Automated pipelines bring in product telemetry, e-commerce, support and marketing data, with quality checks and alerts on each load.
On the gold layer we built a customer 360 view and Power BI dashboards for executives, product and marketing, with self-service access for 50+ business users.
We then trained churn prediction, segmentation and cross-sell recommendation models in Databricks, tracked in MLflow, so customer success sees at-risk accounts before renewal.
Before and after
| Measure | Before | After |
|---|---|---|
| Manual reporting time | 30+ hrs a week | Automated |
| Report generation | Hours | Minutes |
| Pipeline reliability | 92% | 99.5% |
| Customer churn | Baseline | 15% lower |
What's next
With churn scores in production, the next step is L3: an agent that turns each at-risk account into a drafted outreach plan for customer success to review.
