---
name: industrial-ai-pilot
description: Design a measurable, reversible 90-day AI pilot for industrial operations, including baseline, data contracts, safety gates, cost controls and evaluation.
---

# Industrial AI pilot (90 days)

Use this skill when the business has already selected one real process and wants an implementation plan.

## Preconditions

A named process owner, a measurable business problem, available data with legal permission to use, and an accountable human approver.

## Proposed phases

- Weeks 1–2: document process and risk constraints, verify sample data, measure baseline and agree on success/failure.
- Weeks 3–4: build a small read-only prototype; compare with human work without altering production.
- Weeks 5–8: run shadow-mode testing; record accuracy, time saved, errors and exceptions.
- Weeks 9–12: assess cost, security, observability, employee adoption and measurable business value; decide whether to stop or scale.

## Required controls

- Human approval for pricing, contracts, quality release, machine commands and other consequential decisions.
- Versioned data inputs, audit logs, access limits, revocation, backups and rollback.
- Expense caps and explicit timeouts if paid models are ever used.
- Validation against independent examples, including negative cases and failure recovery.
- Clear ownership of post-pilot maintenance and incident response.

## Deliverable

A one-page pilot brief with process scope, RACI, baseline KPI, target KPI, data inventory, system connections, financial assumptions, risks, rollback steps, estimated budget and decision date. Mark unknowns as unknowns; no fabricated ROI or client case studies.