You signed up for an AI subscription in the spring. Since then you have added a note taker for calls, a tool that drafts proposals, an inbox sorter, and a chat window on the website that answers at midnight. The charges land on the first of the month and the work does not feel faster. Your best estimator says the drafts need rewriting, and your office manager has quietly gone back to her spreadsheet.
You did what the articles said to do. Now you have a stack, a bill, and a feeling that the stack is running you.
Why is AI not saving me time?
Because each tool is working on a workflow that has not been drawn. In my experience most of what owners do with these tools is automate a task: the proposal, the inbox, the follow-up. That is fine as far as it goes. It also means the tool polishes whatever step you pointed it at, whether or not that step is the one holding the business back.
Take a man who owns a roofing and restoration company outside Houston, forty people in the busy season. Spring brings hail, and with hail comes a wave of leads. His money lands in two places: insurance carriers, and the property-management firms that run apartment complexes. One firm with a large portfolio can send more work in a week than his ads send in a quarter. He bought a proposal writer, an automated follow-up tool and the midnight chat window, all in one winter.
The proposals went out faster. The jobs did not start sooner. When we drew the lane on one page, from first call to final invoice, the delay sat after the inspection, in documenting the scope so a carrier’s adjuster would agree to it, and in the supplements that followed. His crews were waiting on paperwork. The tool had sped up a step that was not the constraint.
He could not see that because the page did not exist. The tool amplified what he had, and what he had was an undrawn process.
Why do I call this a pilot and mechanic problem?
I tell rooms to think like a pilot before they hire a mechanic, and the roofing owner shows why. The mechanic’s eyes are in the machine. The pilot’s eyes are out the window, on the conditions and the heading. He had spent a winter looking down into the tools while the conditions outside changed. The hail, the carriers and the property managers were the weather. Nothing in his stack was reading it.
What does the pilot do first?
Reads the conditions, then chooses the engine. I use a fixed order: process, then processor, then product. The process is the know-how, how the work gets done. The processor is anything that does the work: a person, a machine, a drone, an AI model. The product is the components you need to deliver.
The order matters because of what is happening to know-how. AI is pushing the price of knowing how toward zero. What holds value when knowledge gets cheap is your process, written down, owned by the business and held in documents rather than in anyone’s head. A subscription can tell a rival roofer how roofing companies generally work. It cannot tell them how yours works. Two other things keep their value, because a subscription cannot buy them: the record of your own customers and your own delivery, and the people who buy from you and work for you because they believe in you.
Private client data stays on machines and networks you control. Outside services are for low-privacy work, or for sensitive work after the names are stripped, and stripping names reduces the risk without removing it. Where the data is regulated, the law and your counsel set the rule.
The tool from the book is plain: codify one lane. Write the steps of one workflow in documents the business owns, date them, and put a named human on the release lever for anything a machine produces.
Once the lane is written, the engine question gets easy. For each step: has the know-how gotten cheaper, and is there now an engine that can run this step behind my process, to my standard? Some quarters the answer is no across the board. The discipline is in asking.
Where should I point the machine?
Outward. The roofing owner took it off his inbox and aimed it at the county: public permit filings, storm reports by street, properties that had just changed hands. That is where the next job sits. Reading it is market work, and the machine does it when the owner tells it where to look.
Automation is the bigger idea. AI is one moment of yes or no. Then the automation does the work. You want the machinery on the lane you drew, doing the repeatable steps, with a person approving anything before it leaves the building.
What can I do this week?
Give it two hours.
- Pick the lane that earns the most, usually first call to final invoice.
- Write every step in plain documents in your own files, not inside one app or one head.
- Date the document. Every later change gets a date too.
- Mark each document High or Low sensitivity. Keep the High ones on systems you control.
- List every automation that touches the lane, and name the person who approves its output before it goes out.
- Circle the one step where the lane stalls. Hold off buying anything for it. Ask what the market is waiting on.
Machines build the road. AI reads the map at speed. You decide where the road goes.
Chapter 41. The Technology Leg of The Method Trap, by John “Jonyprofit” Whaley. One idea, one story, one assignment. About the book
The people in this story are composites drawn from nearly forty years across the table from owners, executives and professionals. Names, industries and numbers are changed.