Diagnose a Sudden Metric Drop with a Ranked List of Likely Causes

Get a ranked table of likely causes for a metric drop, covering tracking, segments, platforms, releases, seasonality and external factors, each with an exact check.

You getSituation read, table and first three actions for the next hour

Prompt

Prompt
Act as a senior product analyst who has led many metric-drop investigations.

Your job: help me diagnose a sudden drop in a product metric and rank the likely causes. Before answering, ask me these questions in one numbered message, then wait for my reply:
1. Which metric dropped, by how much, and compared to what baseline?
2. When did it start (date and hour if known), and is it still falling?
3. Which platforms, regions and key user segments does the product have?
4. What changed recently: releases, experiments, campaigns, pricing, tracking or data pipeline work?

Rules:
- Walk the tree in this order: tracking and data pipeline, segment, platform, release or experiment, seasonality and calendar, external factors.
- Rule out broken tracking first. Do not treat the drop as real until the data is confirmed.
- For every cause, give one exact check: a SQL query sketch, a dashboard cut or a log to inspect, plus the result you would expect if that cause is true.
- Rank causes by likelihood based on my answers, with a one-line reason.
- Keep each check doable in under 30 minutes where possible.

Output format:
1. Situation read (one short paragraph)
2. Table: Rank | Cause | Branch | Why likely | Exact check | Expected result if true
3. First three actions for the next hour
4. Short update to send stakeholders

End with one line naming the missing data that would most change the ranking.

Example output

At a glanceRanked causes of the checkout conversion dropConversion fell from 3.4% to 2.8% after release 4.12
  1. Purchase event not firing in 4.12Rank 1Tracking: new SDK shipped in same release
  2. Checkout bug in 4.12Rank 2Release: drop starts hours after rollout
  3. iOS-only payment failureRank 3Platform: Apple Pay changes in release notes
  4. Paid traffic mix shiftRank 4Segment: new TikTok campaign launched Monday
  5. Post-holiday dipRank 5Seasonality: Monday was a public holiday
  6. Competitor saleRank 6External: no signal yet

Example for: checkout conversion fell from 3.4% to 2.8% on Tuesday at a fictional shop app; iOS, Android and web; app version 4.12 shipped Monday evening.

1. Situation read

Conversion dropped about 18% starting Tuesday morning and has held flat for two days. The timing lines up with release 4.12 and a new analytics SDK. Web traffic is a useful control: if web is stable, the cause is most likely app-side.

2. Ranked causes
RankCauseBranchWhy likelyExact checkExpected result if true
1Purchase event not firing in 4.12TrackingNew SDK shipped in same releaseSELECT app_version, COUNT(*) FROM events WHERE event='purchase_completed' AND date >= '2024-05-13' GROUP BY 1; compare to orders tableOrders table steady, events low on 4.12 only
2Checkout bug in 4.12ReleaseDrop starts hours after rolloutConversion split by app_version for Tue–Wed4.12 converts well below 4.11
3iOS-only payment failurePlatformApple Pay changes in release notesPayment gateway logs: error rate by platformiOS errors up, Android flat
4Paid traffic mix shiftSegmentNew TikTok campaign launched MondaySessions and conversion by utm_sourceLow-intent source share up sharply
5Post-holiday dipSeasonalityMonday was a public holidaySame weekdays last year, week over weekSimilar dip last year
6Competitor saleExternalNo signal yetCheck competitor sites and social feedsVisible sale this week
3. First three actions
  1. Run the rank 1 query against the orders table. If orders are flat, this is a tracking issue, not a revenue issue.
  2. Ask mobile engineering to check crash and payment logs for 4.12.
  3. Pause the staged rollout of 4.12 if version-level conversion is down.
4. Stakeholder update

"Checkout conversion shows an 18% drop since Tuesday. We are checking whether this is a tracking gap from release 4.12 or a real checkout issue. We expect an answer within two hours. No action is needed from you yet."

Missing data: real order counts from the backend would confirm or rule out the top cause.

Written by AI from this prompt. Your results will vary.

Tips and common edits

Why this works

  • Forcing the tracking check first stops teams from chasing a drop that is really a broken event.
  • Each cause gets an exact query plus its expected result, so every check gives a clear yes or no.
  • The fixed tree order covers seasonality and external causes that rushed investigations often skip.

Use this in ChatGPT, Claude or Gemini right after spotting a drop on a dashboard. Have the metric, start time, platform split and a list of recent releases or experiments ready.

Common edits

  • Name your warehouse (BigQuery, Snowflake) so the SQL sketches match your syntax.
  • Swap the stakeholder update for a Slack-length message for a leadership channel.

Prompt details

By
Prompts-GPT editors
Published

Did this prompt work?

Keep exploring

More in Product Management