# Client Pilot Implementation Plan

OpsPilot AI should enter a client environment as a reviewed pilot workflow, not as an autonomous production system. The first implementation goal is to turn one messy operational input into a client-ready brief that a human can approve.

## Review gate

Mike is the review gate at this stage. OpsPilot AI produces draft operational intelligence; Mike reviews the brief, edits weak recommendations, rejects risky output, and approves any client-facing next action.

Nothing generated by OpsPilot AI should be sent to a prospect, customer, CRM, ticketing system, or team workflow without human approval before any client action.

## Phase 1 — synthetic or sanitized pilot

Use synthetic or sanitized examples first. The pilot should prove that the generated brief is useful without creating privacy, confidentiality, or operational risk.

Rules for this phase:

- Use no real customer data.
- Do not enter confidential, client, patient, financial, or personal data.
- Keep the public site read-only.
- Use invite-gated live generation only for approved internal review.
- Capture latency, output quality, missing sections, and reviewer edits.

Expected output:

- workflow summary,
- priority score,
- risk level,
- recommended status,
- missing information,
- next actions,
- pricing/payback rationale when relevant,
- implementation roadmap.

## Phase 2 — workflow mapping

Before connecting to client infrastructure, map the client’s current process.

Discovery questions:

1. Where does the work enter today: website form, email, phone note, CRM, spreadsheet, or ticket?
2. Who reviews the request first?
3. What makes the request urgent, profitable, risky, or unqualified?
4. What information is often missing?
5. What decision should OpsPilot help prepare?
6. Where should the reviewed output go?
7. What are the success metrics for the pilot?

Good success metrics include:

- triage time reduced,
- faster response to high-value opportunities,
- fewer incomplete handoffs,
- more consistent follow-up drafts,
- clearer owner/operator visibility.

## Phase 3 — controlled real-data pilot

Move to real client data only after the client accepts the pilot boundaries and the operational controls are ready.

Minimum controls:

- named users only,
- explicit human reviewer,
- short pilot window,
- data minimization instructions,
- rate/spend limits,
- backup and restore checks,
- no autonomous customer communication,
- no uncontrolled CRM or ticket writes.

At this phase, OpsPilot can process selected real examples, but the output remains a draft for review.

## Phase 4 — infrastructure integration

Infrastructure integration should start with low-friction inputs and reviewed outputs. Writeback and automation come later.

### Integration decision tree

1. **Manual/private review**
   - The operator pastes one request into OpsPilot.
   - OpsPilot returns a draft brief.
   - Mike reviews before sharing or acting.
   - Best for the first client conversation and lowest-risk pilot.

2. **Email or form intake**
   - A website form, forwarding rule, or intake mailbox sends selected requests.
   - OpsPilot generates a brief for review.
   - Output goes to Mike, the owner, or a manager.
   - Good first infrastructure step because it avoids deep system access.

3. **CRM or ticketing writeback**
   - OpsPilot attaches a reviewed summary as a note, task, or priority flag.
   - No customer message is sent automatically.
   - Requires API access, scoped credentials, and a rollback plan.

4. **Automation only after approval**
   - Auto-routing, auto-status changes, and outbound customer responses require explicit approval.
   - Keep human review until the workflow has proven quality, safety, and business value.

## First client implementation package

For each client, create a short pilot brief before implementation:

- client name,
- industry,
- selected workflow,
- intake source,
- output destination,
- reviewer,
- allowed data,
- prohibited data,
- success metrics,
- pilot duration,
- integration level,
- expansion decision point.

## Operating principle

OpsPilot AI should fit into how the business already works. The first win is not deep automation; it is a reliable reviewed brief that helps the operator decide what to do next.