Your automation broke because nobody wrote down the steps

# Your automation broke because nobody wrote down the steps

Most automations that fail did not fail in the tool. They failed before anyone opened the tool, because nobody could say out loud what the process actually was.

This is what that looks like in a real case: a law firm that wanted a representation agreement to go out as soon as a consultation went well, instead of hours or days later.

## The situation

Right now the sequence is manual. Someone reads the meeting notes, decides whether the client wants to proceed, drafts the agreement, and sends it. Sometimes that happens the same afternoon. Sometimes it happens Monday.

What that costs:

- Leads who were ready at 2pm are cooler by Thursday. - Staff re-read the same notes twice to work out what was agreed. - Nobody can tell you how long the gap actually is, because nobody is timing it.

What they wanted instead: the agreement goes out automatically when the consultation shows the client wants to proceed.

## Two things you define before you touch a tool

**The process.** The whole journey, start to end. Booking through to onboarded client.

**The steps.** The individual actions inside it. "Client fills form." "Send agreement." "Update the CRM tag."

You need both, and most people only have the first. They know the journey in their head and assume the steps are obvious. Computers do not make assumptions. They do what you specify and nothing else, and every gap in your description becomes a gap in the automation.

## Level 1: the map anyone can read

Map it visually before building. A whiteboard tool is enough. The one used here was Canva's, and it does not matter which one you pick as long as the whole team can open it.

``` Start → Book consultation → Conduct consultation → Does the client want to proceed? Yes → Send agreement → Signed → Tag as Client No → Nurture sequence End ```

That is the entire firm's intake on one screen. The partner can look at it. The paralegal can look at it and say "that is not what happens, we call them first." That correction is worth more than a week of building.

## Level 2: break each box open

Now take one box and write what actually happens inside it. "Book consultation" becomes:

1. Lead lands on the site or clicks an ad. 2. They submit a form or pick a slot on the calendar. 3. The system adds them as a contact, tags them, and sends a confirmation. 4. If a consultation is booked, they enter a reminder sequence.

Those four lines are the building blocks. Every one of them is something a tool can be told to do. The Level 1 map was not. "conduct consultation" is not an instruction.

## Write it as triggers, or it will not build

The translation step people skip: turn each Level 2 line into a *when this, then that*.

- When a form is submitted → add the contact and tag it. - When a consultation is marked complete → check the meeting notes for whether the client wants to proceed. - When that check passes your threshold → send the agreement for signature.

You cannot say "book a consult, then conduct a consult." That is a description, not logic. Automation is programming in plain sentences, and the sentences have to be exact.

## Where this map is still wrong

Two honest limits, because they will bite in build:

**Reading intent from notes is a judgement call.** A threshold on an intent score is a decision you are handing to software. On a matter where the client is about to sign a legal agreement, decide deliberately whether a human confirms first. There is a version of this where the automation drafts and a person hits send, and for some firms that is the right answer.

**The map above has no exceptions in it.** No unsigned agreement after five days. No missing field. No client who says yes in the meeting and goes quiet. A Level 1 map is meant to be clean; a build is not. When you go to build, every "yes" branch needs a "and if nothing happens" branch beside it.

## What to do with this

Open a whiteboard. Draw Level 1 for one process you run every week. Intake, onboarding, whatever is loudest. Show it to the person who actually does that work and let them correct it. Then break one box into Level 2.

If you cannot get Level 2 written down, you are not ready for the tool, and building anyway is how you end up with an automation nobody trusts and everybody works around.

---

*Written by Daniel Galang, who builds automation systems at Insert Fuel across Zapier, Make and Airtable.*

*If you want a second pair of eyes on your map before you build it: [book a call](/book-a-call).*