The Small AI Habit That Will Give Your Agency an Edge

Most agency owners are using ChatGPT, Claude, or Gemini like Google. Ask a question, take the answer, move on. That's fine for quick stuff. 

But if you're using AI for anything that actually matters, building processes, writing SOPs, designing comp plans, training your team, you're leaving leverage on the table. 

Here is what I learned. Instead of asking your big question, ask the tool to write the prompt first. Then copy that prompt and run it. 

I know. It sounds like an unnecessary extra step that adds time to something you're already using to save time. But here's what's actually happening when you skip it: you're walking into a conversation without knowing what you want out of it. You get an answer, but it's shaped by the tool's assumptions, not yours. You end up editing, re-asking, or just accepting output that's close but not quite right. That back-and-forth costs more time than the 60 seconds you skipped at the start. 

When you ask for the prompt first, you're forced to define what you actually need before the conversation starts. What role should the tool take? What time frame matters? What assumptions should it use? What format do you want back? That extra step clarifies your own thinking before you've typed out a single real question. 

Most of us skip that entirely. We type "create a comp plan for producers" and wonder why the output feels generic. 

Try this instead: "Write a high-quality prompt that will help me design a producer comp plan for a $2M revenue independent agency with a 4-day workweek and a team-based model." 

Now the tool knows what lane it's in. The output reflects that. 

This is worth doing when the task shapes how your agency runs, a new sales process, a training roadmap, a service workflow, a performance scorecard. If you're brainstorming or knocking out a quick draft, don't bother. 

What a good prompt is actually made of 

Once you start writing prompts intentionally, you realize there's a repeatable structure underneath the good ones. It's five parts. 

The first is the role. Tell the tool who it should be. Not "you are a helpful assistant", something specific. "You are a COO of an independent insurance agency with 15 employees and $2M in revenue." That framing changes everything that follows. 

The second is the context. Give it the situation. What's the problem? What's already been tried? What constraints exist? Don't assume it knows your agency. Tell it. 

The third is the task. State exactly what you need produced. Not "help me think through", "write a 90-day onboarding plan" or "draft a producer scorecard with five measurable categories." Specific verbs, specific deliverables. 

The fourth is the constraints. This is the part most people leave out. What should it avoid? What tone? What length? Should it assume you have a small team or a large one? Should it write for agency owners or for staff? The constraints shape the output more than anything else. 

The fifth is the format. Tell it how you want the answer back. A document? A table? A script? Step-by-step? Running prose? If you don't specify, it guesses, and the guess is usually wrong for your situation. 

Put it all together and a prompt that used to be one sentence becomes a paragraph. That paragraph produces something you can use on Monday instead of something you have to rewrite from scratch. 

AI mirrors how clearly you think. Vague instructions produce drifting output. Tight instructions produce something you can actually use. 

Blueprint before building. 

– Michael