# Redesign a process with AI and human review checkpoints

Rebuilds a workflow step by step into automation, AI-drafted, and human-only work, with named review checkpoints, accountability, and risk controls.

- Topic: Working with AI
- Framework: Few-Shot
- Level: Advanced
- Tags: Analyze, Fill-in Template
- Source: https://www.thinkwithprompt.com/prompts/redesign-a-process-with-ai-and-human-review-checkpoints

## Prompt

```text
Redesign this process so automation, AI, and people each handle the steps they suit best, with clear human review checkpoints. I already know AI could help here; I need the redesigned workflow.

**Process:** [NAME and purpose]
**Current steps:** [PASTE steps with who does them, systems used, and time each takes]
**Tools available:** [e.g. Microsoft 365 Copilot, ChatGPT Enterprise, Power Automate, Zapier, our CRM]
**Data involved:** [types of data - customer, financial, personal, confidential]
**Company AI policy or limits:** [OPTIONAL - what is approved]
**Who is accountable for the final output:** [ROLE - e.g. "AP team lead"]

For each step, decide the best mode:
- Automate: rule-based, repetitive, structured data, no judgment (e.g. moving data between systems, reminders, routing).
- AI executes, human spot-checks: low-risk, easy to verify, reversible (e.g. first-draft summaries, tagging, formatting).
- AI drafts, human decides: needs judgment, but AI can speed up preparation (e.g. analysis, recommendations, customer replies).
- Human only: high stakes, accountability, empathy, ethics, legal effect, or decisions about people.

Explain each choice using: consequence of an error, how easy errors are to spot, reversibility, data sensitivity, and whether a person must be accountable.

Return:
1. A table: Step | Current owner | Mode | Tool | Why | Human checkpoint | Error risk.
2. The redesigned workflow as numbered steps, showing handoffs between AI, automation, and people.
3. Where human judgment is required: for each checkpoint, who reviews, what they must check, and what triggers sending work back.
4. Estimated time saved per run, using my times only; mark estimates [EST].
5. Risks: hallucination, data leakage, over-trust, bias, and the controls for each.
6. A small pilot to test the design.

Don't put confidential or personal data into tools not approved by my company. People remain accountable for decisions; AI output must be reviewed before it reaches customers or affects employees.
```

## What to fill in

- NAME and purpose
- types of data - customer, financial, personal, confidential
- OPTIONAL - what is approved
- ROLE - e.g. "AP team lead"
- EST

Free to copy and use. From ThinkWithPrompt (https://www.thinkwithprompt.com/), a library of 2,663 curated AI prompts.