thinkhard.SI

Prompt repository Operations

Find the few causes behind most defects, errors or returns

You have a log of defects, errors or returns and want to know which few causes account for most of them, so you fix those first.

Registry no.
OPS-019
Collection
Operations
Laid out for
Any chat model
Score
10 out of 10
Blanks to fill
5
Added

OperationsAny chat

The prompt

Score 10 of 10, Excellent

Use it when You have a log of defects, errors or returns and want to know which few causes account for most of them, so you fix those first.

  • {{audience}}
  • {{period}}
  • {{defect_log}}
  • {{volume}}
  • {{categories}}
Full prompt and why it works
Prompt
Analyze the log below and find which causes account for most of the problems, for {{audience}} to decide what to fix first. I need a ranked count of causes with the share each makes up (a Pareto analysis), not a general list of quality tips.

Defect, error or return log for {{period}} (date, product or process, defect type or reason, quantity, cost if known, notes):
"""
{{defect_log}}
"""

Total units produced, orders shipped or transactions processed in the period: {{volume}}
Our standard list of defect types, if we have one: {{categories}}

Steps:
1. Group the entries into defect types. Use our list if given. If the log uses free text, group similar descriptions and show the grouping as a table with the columns Group, Raw descriptions, Entries, so it can be checked.
2. Count each type by quantity, and by cost where costs are given. If an entry has no quantity, count it as 1 and say how many you counted that way. If only some entries have a cost, total cost only for those and say how many lack one.
3. Rank the types and add a running percentage.

Produce:
1. A table with the columns Defect type, Count, % of total, Running %, Cost (if known). Keep "Unclassified" as the last row whatever its size, and say if it is over 10% of the total.
2. The vital few: the types that make up about 80% of the total, in one sentence, for example "Three of 11 defect types (wrong labels, short picks and crushed corners) make up 78% of returns." If no small group reaches about 80%, say so plainly rather than forcing one.
3. Overall defect rate: total defects / volume, with the calculation, only if both are in the same unit (units against units, orders against orders); otherwise say what figure is needed.
4. Patterns worth checking: up to 5, such as a type concentrated on one line, shift, product or week, each with the counts behind it.
5. Next questions: what to look at to find the root cause of the top types.

Rules:
- Count only what the log shows, and make sure the counts add up to the total quantity in the log.
- Do not state causes behind the defect types. Patterns are leads to check, not conclusions.
- Keep the text outside the tables under 250 words.

If an entry is unclear or could belong to two groups, put it in "Unclassified" rather than forcing it into a group.

Variables

{{audience}}
Who decides what to fix, for example "the quality lead and production manager".
{{period}}
The dates the log covers.
{{defect_log}}
One line per defect, error or return, with date, product or process, type or reason, quantity and cost if known.
{{volume}}
Total units, orders or transactions in the period, saying which, so a rate can be worked out.
{{categories}}
Your standard defect or reason codes, or "none, group them for me".

Why it works

  • What to make. It asks for a ranked count with a running percentage, the specific analysis that shows where to start, instead of generic quality advice.
  • What good looks like. Free-text groupings show which raw descriptions went into each group, and counts must add up to the log, so anyone can challenge how the count was built.
  • If unsure. If no small group of causes really dominates, it says so instead of forcing an 80/20 story, and unclear entries stay visible in Unclassified.

How to use it

  1. Press Fill in variables and type your details into each blank. What you type stays in this browser.
  2. Copy the filled prompt into the AI tool you already use, such as ChatGPT, Claude or Gemini, or open it in the Grader or the Prompt Tester here.
  3. Before you rely on it, try it on three or four real examples. The Prompt Tester runs them all at once and shows which ones go wrong.
Use it from code

Get this prompt as JSON from the free REST API, with your own text in the blanks, or let an AI assistant fetch it through the MCP server. No key.

curl -s https://thinkhard.si/api/v1/prompts/defect-pareto-analysis

With the MCP server added to Claude, ChatGPT, Cursor or VS Code, call get_prompt with {"prompt": "defect-pareto-analysis"}, or just ask for it by name.