At first glance, activation did not look like the main problem. The overall activation rate was 50.92%, which made performance look relatively stable on the surface. But that average turned out to be doing too much hiding. Once the data was segmented, it became clear that activation quality was not evenly distributed across the subscription base. Some source and plan combinations were producing much stronger starts than others, while weaker combinations were getting flattened into the same headline number. The real issue was not the average itself. It was the variation underneath it.

Business problem

The real business risk was not low activation in general, but misreading where activation strength and weakness were actually concentrated. A stable headline rate can make performance look more consistent than it is, which makes it easier to miss the subscription segments where friction is quietly building. In this case, the important question was whether activation quality changed depending on how subscriptions were acquired and which plan tier they entered. That is the difference between reporting activation and diagnosing it.

The analysis had to be built at the subscription level, not the account level, because account-level rollups flattened too much of the variation that mattered. I used a multi-table SaaS subscription dataset from Kaggle, structured across account, subscription, and feature-usage tables, as a controlled environment for testing activation logic and segmentation. The first task was to inspect the schema, validate the joins, and test whether activation could be measured through usage timing. That logic did not hold up cleanly: for a large share of subscriptions, first usage appeared earlier than subscription start, which made a simple time-to-activation definition unreliable. Rather than forcing a weak method, I shifted to a usage-depth approach and defined activation using a stricter behavioral threshold of at least five usage records and five distinct features used. From there, I built a final subscription-level activation table in Python and used Tableau to diagnose how activation quality changed across source and plan segments.

Findings

The overall activation rate came out to 50.92%, but the referral-source view made the variation underneath that average much easier to see. Trial status added little separation, while referral source revealed a clearer activation-quality gap: organic and event-driven subscriptions consistently outperformed ads and other sources. Plan tier also changed the picture, but not in the expected direction. Enterprise underperformed both Basic and Pro, suggesting that higher-value subscriptions were not automatically producing stronger starts. The strongest insight came from the combined view. Activation quality was not driven by source alone or tier alone, but by the specific source-tier environments subscriptions entered through.

Business interpretation / implication

The analysis became more valuable once activation was treated as a diagnostic problem rather than a single KPI. The average alone could not show where performance was holding up and where it was weakening early, but the segmented view could. That shift makes the output more actionable. Instead of reacting to one blended activation number, the business can investigate weaker source-tier combinations directly and ask whether the friction is coming from acquisition fit, onboarding burden, plan complexity, or some interaction across all three. That is the practical advantage of diagnosing activation quality instead of only reporting it.

key outcome

The value of the project was not the headline activation rate itself, but what happened once that number was broken apart. A stable-looking KPI became a much more useful diagnostic once the subscriptions were segmented by source and plan tier. Python was used to test the method, reject weak timing logic, and build the final working table, while Tableau made the variation visible in a way the average could not. The final takeaway was that activation did not need a louder KPI. It needed a better lens.