Workflow Audit
Most AI gets bolted onto a broken process and makes the chaos faster. This skill maps a workflow end to end, finds the actual constraint, and decides where AI belongs — and where it does not.
When to use this skill
Use this skill when:
- A recurring workflow is slow, inconsistent, or error-prone.
- You are tempted to "add AI" but are not sure where it would help.
- You want a defensible before/after baseline for an improvement.
Inputs needed
- The workflow name and its trigger (what starts it) and outcome (what ends it).
- The current steps, owners, and tools, even if rough.
- Rough time and error data per step, if available.
- The volume: how often the workflow runs.
Process
- Define the workflow's trigger and its done-condition in one line each.
- List every step with its owner, tool, typical time, and rework rate.
- Mark each step as value-adding, necessary-non-value, or waste.
- Find the bottleneck: the step that constrains the whole workflow's throughput.
- For the bottleneck and the top waste steps, ask whether the fix is process, tooling, or AI.
- Propose the smallest change that relieves the bottleneck; ignore the rest for now.
- Define the baseline metric and the target so the change can be measured.
Prompt or workflow
You are auditing a recurring business workflow.
Workflow: [NAME]
Trigger: [WHAT STARTS IT]
Done when: [WHAT ENDS IT]
Runs: [HOW OFTEN]
Current steps (owner, tool, time, rework):
"""
[PASTE STEPS]
"""
Produce:
1. MAP: a numbered step list with Owner | Tool | Time | Rework for each.
2. CLASSIFICATION: tag each step value-add / necessary / waste.
3. BOTTLENECK: name the single step that constrains throughput, with evidence.
4. FIX TYPE: for the bottleneck and top 2 waste steps, classify the fix as
process / tooling / AI, and say why.
5. SMALLEST CHANGE: the one change to make first.
6. MEASUREMENT: baseline metric + target to prove it worked.
Rules:
- Identify exactly one bottleneck.
- Do not recommend AI where a process or tooling fix is cheaper.
- Prefer the smallest viable change.
Quality checklist
- Trigger and done-condition are both explicit.
- Every step has an owner and a tool.
- Exactly one bottleneck is identified, with evidence.
- Each proposed fix is classified as process, tooling, or AI.
- The first change is the smallest one that relieves the bottleneck.
- A baseline metric and target are defined.
Common mistakes
- Optimizing a step that is not the bottleneck.
- Recommending AI for a problem a checklist would solve.
- Skipping the baseline, making improvement impossible to prove.
Example output
Workflow: Monthly close. Trigger: month-end. Done when: books locked.
Bottleneck: intercompany reconciliation (avg 3 days, 40% rework).
Fix type: tooling first (auto-match), AI second (exception narratives).
Smallest change: auto-match rule for the top 3 recurring intercompany entries.
Measurement: baseline 3 days -> target 1 day for reconciliation.
Related skills
- Prompt Refinement Loop — once AI is the right fix, make the prompt reliable.
- Meeting Notes to Action Plan — to turn audit findings into owned actions.
- Decision Memo Builder — to get sign-off on the proposed change.
Attribution
This skill was created by Vectory and is licensed under CC BY 4.0.
Source: https://vectory.io/skills/workflow-audit
Attribution: "Workflow Audit" by Vectory.