Diagnose a KPI Drop: Junior Analyst Root Cause Session

Who This Helps

This is for junior analysts who get a Slack ping that a KPI dropped 12% overnight. Your boss wants answers by end of day. You want to deliver a clean analysis with clear recommendations, not a messy spreadsheet.

Mini Case

Mei, a junior analyst at a subscription company, saw the weekly active users drop 12% in one day. She used the Data Reliability Leadership course's Incident Triage mission to run a calm, structured first 30 minutes. She found the root cause: a failed data pipeline for the mobile app. She shipped a fix and a recommendation to add an alert. Her team avoided a full incident.

Do This Now (5 Steps)

1. Check the metric definition. Open the data contract for that KPI. Confirm what it measures and what sources feed it. If no contract exists, note that as a risk.
2. Look at the time window. Compare the drop to the same hour yesterday, same day last week, and same day last month. A 12% drop might be a seasonal pattern.
3. Slice by dimensions. Break the metric by platform, region, or user segment. Mei found the drop was only on iOS. That narrowed the search.
4. Check data pipeline health. Look at the monitoring dashboard for that data source. If the pipeline failed, you'll see a gap or error. Mei saw a failed job at 2 AM.
5. Write a one-page summary. State the root cause, the impact (12% drop, 7 days of lost data), and one clear recommendation (add an alert for that pipeline).

Avoid These Traps

• Don't panic and guess. A 12% drop feels big, but guessing wastes time. Follow the steps.
• Don't skip the data contract. If you don't know what the metric means, you can't diagnose it.
• Don't blame the data team. Focus on the process, not people. The pipeline failed; fix the pipeline.
• Don't send a raw query. Your boss wants a story, not SQL. Summarize.
• Don't forget the recommendation. A diagnosis without a fix is just a complaint.
• Don't overcomplicate. One root cause, one recommendation. That's it.
• Don't ignore the time window. A 12% drop in one hour is different from a 12% drop over a week.
• Don't skip the alert. If you found a pipeline failure, recommend a monitor.

Your Win by Friday

Ship a one-page analysis that says: "The 12% drop in weekly active users was caused by a failed data pipeline for iOS. I recommend adding an alert for that pipeline. Here's the fix." Your boss will trust your work. You'll feel like a detective who cracked the case. And you'll have a template for the next drop. (There's always a next drop.)