---
name: manufacturing-ai-diagnostic
description: Diagnose candidate opportunities for AI in an industrial company, using only verified public facts or company-provided information, with human approval for production decisions.
---

# Manufacturing AI diagnostic

Use this skill when asked to identify possible AI projects for a manufacturing company. This is a methodology, not permission to access private systems.

## Inputs

1. Company and industry, with links to authoritative sources.
2. Process owner and the business problem.
3. Existing data sources, volume, update cadence, retention and access permissions.
4. Baseline of current cost, speed, quality or exceptions.
5. Safety, personal-data, commercial-confidentiality and regulatory constraints.

## Method

1. Separate independently verified facts, user-supplied assertions and hypotheses.
2. Map one process as it currently operates. Avoid a company-wide transformation as the first project.
3. Name at most three candidate AI use cases. For each give: why, data requirements, integration dependencies, output, failure modes, KPIs, human review gate.
4. Prioritize reversible read-only or recommendation tasks over operational writes.
5. Select one small pilot with baseline metrics and an explicit rollback/stop criterion.
6. Ask targeted questions for missing evidence instead of inventing economics.

## Format

Use a concise table: process / opportunity / data / KPI / human control. Include citations to source URLs, assumptions, and limitations. Distinguish illustrative KPIs from measured results.

## Hard rules

Never claim to have visited a company or accessed operational data without evidence. Do not promise ROI or production safety. Do not autonomously send mail, alter ERP data, control production, approve quality release or publish company-specific content.