Workflow Automation Patterns

2.1 Identifying Automation Opportunities

Not all workflows are equally automatable. Some are too complex, too dependent on human judgment, or too risky to automate. Your job is to identify the highest-value automation opportunities—those that deliver significant impact with manageable risk.

Why this matters: Automating a workflow that saves 30 minutes per week yields modest value. Automating a workflow that saves 10 hours per week, improves consistency, and frees your team to do more strategic work is transformational. Identifying which is which requires discipline.

Categorizing Automation Opportunities

• High-Potential Automation: Structured tasks (inputs are clear, process is rule-based, outputs are defined), repetitive (happening weekly or more), and low judgment (decision criteria are explicit, not intuition-based). Example: extracting metadata from documents, categorizing customer inquiries, generating routine reports.
• Medium-Potential Automation: Some structure and some judgment mixed. May require sampling human review. Example: generating first-draft analyses, creating summary documents, preliminary research synthesis.
• Low-Potential Automation: High judgment required (decisions depend on context, nuance, strategic insight), infrequent, or high-stakes (errors are costly). Example: strategy decisions, client relationship management, complex contract negotiation.
• Deferred Automation: Currently low-potential but worth revisiting. As AI improves or your team gains skill, this may become automatable. Example: legal opinion writing (improving but still requires attorney expertise).

Prioritization Framework

Once you have identified opportunities, prioritize by impact and risk:

1. High impact, low risk: Do these first. Quick wins that prove AI value.
2. High impact, medium risk: Do these second. Design robust quality controls and pilot with a subset of work.
3. Medium impact, low risk: Do these in parallel. They are easy and contribute to overall capability.
4. High impact, high risk: Do these last and only after you have experience with lower-risk automation.

2.2 Designing Multi-Step AI Workflows

A multi-step workflow is an orchestrated sequence where each step transforms the input and passes it to the next. Think of it like an assembly line: each station does its part, then passes the work to the next station.

Why this matters: Instead of a human manually running a prompt, reading the output, copying pieces to another prompt, running it again, and so on, the workflow does all this automatically. The human focuses on oversight and quality assurance, not mechanical work.

Designing Workflows: Sequential Process

1. Define the end goal: What is the final work product? Be specific. Not just "analysis," but "executive summary, 3–5 paragraphs, highlighting risks and opportunities."
2. Work backward to identify steps: What needs to happen before the final step? What before that? Continue until you reach your starting point (the raw input).
3. For each step, define inputs and outputs: What information does this step need? What exactly does it produce? This clarity prevents downstream surprises.
4. Design decision points: Where does the workflow check quality or escalate to human review? A workflow should never assume AI output is perfect. Build in verification.
5. Plan error handling: If one step fails or produces unacceptable output, what happens? Restart? Escalate? Route to a different path?
6. Test end-to-end: Don't test steps in isolation. Run the full workflow on real data. Do the outputs meet your requirements? Does anything break?

Workflow Design Prompt

Use this prompt to design a workflow for your specific use case: