AI Prompt to Audit a Workflow Before Automating It

PROMPT GUIDE / AI AUTOMATION

Use this prompt before automating a workflow with AI. It turns a vague automation idea into an evidence-based readiness audit, exposes fragile steps, and defines where human approval must remain.

Act as a workflow analyst. Audit the workflow below before recommending any AI automation.

WORKFLOW: [DESCRIBE THE CURRENT WORKFLOW]
GOAL: [DESIRED BUSINESS RESULT]
USERS: [WHO PERFORMS OR DEPENDS ON IT]
TOOLS AND DATA: [SYSTEMS, DOCUMENTS, AND DATA SOURCES]
CONSTRAINTS: [SECURITY, PRIVACY, BUDGET, COMPLIANCE, OR TIMING]

Return these sections:
1. Current workflow map: inputs, steps, decisions, handoffs, outputs, and owners.
2. Evidence gaps: facts that must be confirmed before automation.
3. Automation candidates: repeatable, rules-based steps with stable inputs and measurable outputs.
4. Keep-human steps: judgment, empathy, accountability, approval, exception handling, or sensitive decisions.
5. Risk register: failure mode, likelihood, impact, detection method, and mitigation.
6. Data readiness: required fields, quality problems, permissions, retention, and confidential information.
7. Proposed pilot: the smallest reversible test, success metrics, sample size, monitoring, and rollback trigger.
8. Recommendation: automate now, redesign first, assist only, or do not automate—supported by evidence.

Rules:
- Do not assume the workflow is correct because it already exists.
- Do not invent process steps, performance data, legal requirements, or tool capabilities.
- Mark every unverified statement as an assumption.
- Prefer human-in-the-loop assistance when errors could materially affect a person, customer, payment, contract, or regulated decision.
- Separate efficiency gains from quality, safety, and accountability.
- Ask up to seven targeted questions if essential information is missing.

How to use this prompt

1. Describe the real workflow. Include workarounds, handoffs, exceptions, and rework—not only the official process.
2. Add representative inputs. Provide anonymized examples so the analysis reflects actual variation.
3. Review assumptions. Confirm owners, data access, error costs, and approval requirements with the people doing the work.
4. Run a reversible pilot. Test one narrow step before connecting the entire workflow.

What this prompt helps you do

• Find automation opportunities that are genuinely repeatable.
• Prevent a broken process from becoming a faster broken process.
• Keep human judgment where consequences, ambiguity, or empathy matter.
• Define measurable success and rollback conditions before implementation.
• Create a practical risk register for stakeholder review.

When not to use it

Do not use this prompt as a substitute for security review, legal advice, compliance approval, employee consultation, or professional risk assessment. It structures the analysis; accountable people still make the decision.

Example decision

A team may automate extraction of structured fields from standardized forms, while keeping contract interpretation, customer exceptions, payment approval, and final sign-off with authorized people. The best result is often assisted work—not full autonomy.

Frequently Asked Questions

Should every repetitive task be automated?

No. Repetition alone is not enough. Inputs must be sufficiently stable, errors detectable, and the value of automation greater than the cost of monitoring and exceptions.

What if the workflow is poorly documented?

Map the real workflow first. Automation should follow process discovery, not replace it.

Can this prompt choose an AI tool?

It can define requirements and evaluation criteria, but tool selection should use verified capabilities, security terms, total cost, and a controlled trial.

Read the full decision framework: What to Automate With AI—and What to Keep Human