Before any of this works, somebody has to write down how the work currently gets done.
Not how it is supposed to get done. Not the version in the operations manual nobody has opened since 2019. How it actually happens, including the part where you text yourself a reminder and the part where Maria just knows.
This is the least popular thing we ask for and the most reliable predictor of whether a project succeeds.
Why it is not optional
A process that lives in your head has a quality that makes it invisible to you: it adapts. You handle the weird cases without noticing you are handling them. You skip a step when it obviously does not apply. You call the customer instead when something feels off.
You cannot hand that to anything. It has to become explicit before it can become automatic, and the act of making it explicit is where you discover that four people do it four ways and two of the steps exist because of a problem you solved in 2021 that no longer exists.
Most owners find real money in this exercise before a single tool gets installed. Not because writing things down is magic. Because nobody had ever looked at the whole thing end to end.
The exercise
Twenty minutes. Paper is fine. Pick one process that annoys you — not the biggest one, the most annoying one.
1. What starts it? Something happens in the world. A phone rings. A form comes in. A job gets marked complete. Be specific about the trigger, because that is what a system will eventually watch for.
2. What happens next, in order? Every step. Including the ones that feel too small to write. Look up the address. Check if we've been there before. Text Dave to see if he's free Thursday. The small steps are usually most of the time.
3. Who does each step? Name a person. If the answer is "whoever sees it first," write that down exactly, because that is a real finding and it is probably why things fall through.
4. Where does it stall? Be honest. There is one step where things sit. There always is. Mark it.
5. What happens when it goes wrong? Not the happy path. The customer is not home, the part is backordered, the price changed. Write down the three most common ones.
That is the whole exercise.
What you will find
Three things, almost every time.
A step that exists for no current reason. Some form that gets filled out and filed and never read. Delete it. Free.
A stall that is one person deep. Everything waits on the owner to approve something, or on the one person who knows the pricing. This is your actual constraint, and it is usually not the thing you thought was the problem.
A step that only happens when someone remembers. The follow-up. The confirmation. The reminder. This is the automation candidate — not the step that takes the longest, the step that gets skipped.
That last distinction matters more than anything else here. People assume you should automate the slowest step. Usually you should automate the least reliable one. A step that takes ten minutes and always happens is costing you ten minutes. A step that takes one minute and happens half the time is costing you jobs.
What to do with it
Nothing yet.
Sit with it for a week. Cross out the steps that should not exist. Fix the ones that need a decision rather than a tool — a lot of stalls are a pricing rule nobody wrote down, not a software gap.
Then look at what is left and ask which single step, if it happened every time without anyone remembering, would change your month.
That is the first thing to build. One step. Not the whole process.
Why this is the whole method
Every failed AI project we have been called in to rescue skipped this. Somebody bought a tool for a problem described in a sentence — we need to follow up better — and then discovered during setup that nobody could agree on what following up meant, when it should happen, or who owned it.
At that point you are paying a subscription to have an argument you could have had for free on a legal pad.
Twenty minutes. Before you spend anything.