You've read enough. You know ChatGPT and GPT-4 exist. You know they're useful. But you don't know how to actually build something your team uses regularly.
Here's the step-by-step guide. This doesn't require coding. It requires clear thinking.
Step 1: Identify the Problem You're Solving
Pick one thing your team does repeatedly that's boring but necessary. Not something innovative. Something that sucks up time.
Examples: Initial client intake questions. Email follow-ups after meetings. Contract summaries. Research briefs. FAQ responses.
Pick something your team does at least once a week. Something that's currently taking 30 minutes to an hour per instance.
Don't pick something that requires legal judgment or critical analysis. Pick something structured and repeatable.
Step 2: Map Out the Current Process
Write down exactly what happens right now:
- What does the person start with? (Email, form, request, information)
- What steps do they follow?
- What is the output?
- Who reviews it?
- What modifications usually happen?
This doesn't need to be perfect. Just accurate enough to understand the flow.
Step 3: Design the AI-Assisted Version
Now redesign it with AI in the loop:
- Same input
- Feed the input to ChatGPT or GPT-4 with a clear prompt
- AI generates a draft output
- Person reviews and edits (usually takes 5-10 minutes)
- Same output as before
That's it. You're not replacing the person. You're replacing the 30 minutes of drafting with 30 seconds of AI generation + 5 minutes of review.
Step 4: Write the Prompt
This is the critical part. Your prompt tells the AI exactly what to do.
Good prompt structure:
- Context: "You are helping a [law firm / clinic / consulting company] with [task]"
- Input: "The input will be [description of what you'll paste]"
- Task: "Generate [specific output]"
- Format: "Format as [bullet points / paragraphs / structured list]"
- Tone: "The tone should be [professional / friendly / formal]"
- Constraints: "Do not [make assumptions / invent information / go beyond X length]"
Example: "You are helping a law firm with client intake. The input will be a free-form client questionnaire. Generate a summary of key facts about the client's legal situation in 3-4 bullet points. Focus on what's relevant to corporate law matters. Do not assume anything not explicitly stated in the questionnaire."
Step 5: Test With Real Work
Don't test with made-up examples. Test with actual work from your firm.
Pick 5-10 instances of the task you've chosen. Run them through your AI + human workflow.
Measure:
- How long does AI generation take? (Usually seconds)
- How long does human review take?
- How often does the AI get it basically right vs. needing major rework?
- Are there any misses or errors that concern you?
- Total time compared to doing it manually?
If you're saving 50%+ of the time, you've got something worth scaling.
Step 6: Refine the Prompt
After testing, you'll probably see ways to improve the prompt. Maybe the AI is including things you don't need. Maybe it's missing something important.
Adjust the prompt. Test again. Iterate a couple of times.
You're looking for "pretty good consistently" not "perfect occasionally."
Step 7: Train Your Team
Once you have a working workflow, show your team.
"Here's the new way we're doing [task]. It works like this: [explain the workflow]. It should save you about X minutes per instance. Here's how to use it. Here's what to check when you review the output."
Let them try it on a few tasks while you're available to help.
Step 8: Document It and Measure Impact
Write down the prompt and the process. Save it somewhere your team can access it.
Track how much time is actually being saved over two weeks. If it's close to what you predicted, you're done. If it's less, figure out why and adjust.
Step 9: Identify Your Next Workflow
Now do it again with the next task.
Each workflow you build gets easier. By the third one, your team understands how to think about this.
The Real Point
This isn't complicated. It's:
- Pick a problem
- Write a prompt
- Test with real work
- Train your team
- Repeat
You don't need special technology. You don't need consultants (unless you want them). You just need to think clearly about what your team does and how AI can help them do it faster.
That's how you build your first AI workflow.