# Dhi Industrial AI Pilot Checklist

Use this checklist before green-lighting a pilot for transit, logistics, warehouse, or industrial safety monitoring.

## 1. Define the first incident class
- Pick one workflow to validate first: platform edge, track trespass, forklift conflict, fall detection, or smoke/fire.
- Define the real failure mode you are trying to catch sooner than your current process.
- Write down the operator who owns the alert and the action they are expected to take.

## 2. Confirm camera fit
- List the exact cameras that cover the pilot zone.
- Confirm RTSP or ONVIF access for each stream.
- Note blind corners, night conditions, glare, occlusion, and any current VMS alarm dependencies.

## 3. Confirm deployment constraints
- Document whether raw footage must remain on-premises.
- Confirm whether the environment can host Jetson, AGX Orin, or industrial x86 compute.
- Note network limits, retention policies, and whether alerts must continue during WAN outages.

## 4. Confirm integration requirements
- List the VMS in use today.
- Define where alerts need to land: native VMS alarm, email, webhook, SMS, operator dashboard, or dispatch workflow.
- Confirm whether policy-approved clips can be exported or only metadata can leave the site.

## 5. Define measurable success
- Target alert latency.
- False-positive threshold.
- Detection coverage threshold.
- Incident-discovery-time improvement.
- Near-miss visibility or response-time improvement.

## 6. Set the pilot decision gate
- Decide how long the pilot runs.
- Decide who signs off on success.
- Decide what evidence is required to expand beyond the first zone or first site.

## 7. Bring to the first review
- Camera map or sample floor plan.
- VMS and camera inventory.
- Privacy or union-review constraints.
- The one workflow you want to improve first.