Almost every conversation I have about AI starts with an apology. Some version of: I know we should have our processes documented, but things are a mess right now. I have heard it from accountants, photographers, agency owners, therapists, and a wedding venue operator running three businesses off one calendar.
Here is what I say back. The mess is not the obstacle. The mess is where your hours are going, which makes it the only sensible place to start.
The documentation-first advice is backwards
The standard move is to hire someone to write SOPs before you touch AI. Six months later you have a binder nobody opens and a team that has already invented its own workarounds.
The businesses I have seen get real value from automation did it in the opposite order. They took one ugly, repeating workflow, recorded it, and pointed a model at it. The SOP showed up on the other end, written in plain language, matching what actually happens instead of what someone wished happened.
Order of operations matters. Documentation is a byproduct of automation, not a prerequisite for it. If you wait until your business is clean enough to automate, you will wait forever, because the mess regenerates faster than you can write it down.
Find the workflow you would be embarrassed to show anyone
You already know which one it is. It is the thing you do at 10pm that you would never put on a landing page.
Three filters help me narrow it down:
- It repeats at least weekly.
- You are the only person who knows how to do it.
- Nobody would pay you to describe it out loud.
Common winners in founder-led service businesses: proposals and scoping documents, client onboarding, weekly client reporting, invoice follow-up, and pulling together the welcome packet for a retreat or event.
Two concrete steps. First, run Toggl or RescueTime for one week so you have actual numbers instead of a feeling. Second, keep a running note titled "stuff only I can do" and add to it every time you catch yourself doing the same thing twice. That note becomes your automation roadmap.
Record yourself doing it once, badly
Do not write the process. Perform it.
Open Loom or QuickTime, share your screen, and walk through one real instance out loud. Real client, real numbers, no rehearsal. Talk through the stumbles. "Oh wait, I need to check whether the contract is signed first" is a genuine step, and it is exactly the kind of thing you would have left out of a written SOP because it feels obvious to you.
Twenty minutes of recording beats three hours of trying to write clean instructions. Otter or Fathom will transcribe it if you want text. Then do it twice more with different clients, because three real examples are what turn a recording into a usable pattern.
Build a draft machine, not an autopilot
This is where most people overshoot. They try to build a system that sends the proposal. Do not do that. Build a system that hands you a draft you can fix in four minutes.
A build that works: a Tally or Google Forms intake form, Zapier or Make as the plumbing, a Claude Project or ChatGPT custom GPT loaded with ten examples of your best past work, and Airtable or a Gmail drafts folder as the landing spot. The client fills the form, the model drafts the scope using your examples as the voice reference, the draft appears in a queue, and you edit.
The first proposal I ran through this took 45 minutes to write by hand. The draft version takes me about nine minutes of editing. At six proposals a week, that is most of a workday back, and the quality went up because the model never forgets to include the revision policy.
Full agentic pipelines sound wonderful in a demo and terrify me during client onboarding. Drafts are where the value is. That's the whole point.
Automate the approval, not the judgment
Every workflow needs one place where a human says yes. Usually that is a Slack approval button or a drafts folder you check once in the morning.
Some things should never be fully automated in a founder-led service business: final pricing on custom work, apology emails when something goes wrong, and anything that ends a relationship. Those moments are where being a person is the actual product. Everything around them can be machinery.
The 30-day messy starter
Week one: pick one workflow and time it honestly.
Week two: record three real instances, twenty minutes each.
Week three: build the draft machine. Form, Zapier, model, queue.
Week four: measure. Compare the minutes against your baseline. If it does not beat the old way, kill it and pick a different workflow. That is a real result, not a failure.
One rule I hold to: do not start a second workflow until the first one saves you three hours a week. The founders who scale this stuff are not the ones with the most tools. They are the ones who finish one thing.
If you want to see the recording and the first build happen live, the workshop at workshop.mastermindshq.business walks through both, step by step.
I have trained more than 90,000 people at this point, and the pattern never changes. The people who wait for clean systems stay busy forever. The people who start with the mess get their time back.
If this resonates and you want to build this kind of infrastructure in your own business, the Mastermind is where we do the work live. You can learn more at mastermindshq.business.
