Anomaly Detection & Root Cause

Generates a ranked list of likely root causes for a sudden metric spike or drop, with a step-by-step investigation plan.

Best For

•Solo founders who notice a conversion drop and don't know where to start looking without a data team to investigate.
•Indie SaaS operators diagnosing a signup or revenue anomaly who want a prioritized checklist instead of guessing randomly.
•Freelance growth consultants troubleshooting a client's metric dip who need a structured, defensible investigation process.

Prompt Template

You are a data analyst investigating a metric anomaly. I'm going to describe a sudden spike or drop in a business metric. Help me identify the most likely root causes. Output: 1. Anomaly Summary: what changed, by how much, and the time window 2. Hypothesis List: rank 5-7 possible root causes from most to least likely, each with: hypothesis, likelihood (High/Medium/Low), evidence needed, how to check 3. Investigation Playbook: step-by-step order to investigate, fastest/cheapest hypotheses first 4. Quick Fixes: if any hypothesis is confirmed, the immediate mitigation 5. Prevention: what monitoring/alerting would catch this faster next time The anomaly: [DESCRIBE: what metric, normal range, what happened, when, any known context like deploys/campaigns/external events]

Pro Tip

Mention every recent change you can think of, even ones that seem unrelated like a new ad campaign or a minor deploy, since the ranked hypothesis list depends entirely on what context you provide.

Example Output

A sample of what this prompt produces once you fill in the placeholders.

An investigation ranking 'new Google Ads campaign sending low-intent traffic' and 'broken signup form' as the two highest-likelihood causes for a conversion rate drop, with a playbook checking the form for JS errors first since it takes only two minutes to rule out.

Recommended Tools

Claude

ChatGPT

How to use this prompt

Copy the prompt template using the button above.
Paste it into your preferred AI assistant (ChatGPT, Claude, Gemini, etc.).
Replace all bracketed placeholders like [TOPIC] with your specific details.
Send the prompt and refine the output as needed.