What Should I Automate First in My Business?

---

The story

Someone asked me to automate their quoting process. They hated it. Four hours a week, tedious, error-prone, obviously a machine’s job. Straightforward.

So I asked them to walk me through it, and about ninety seconds in, it fell apart.

“Well, it depends on the client.” Which clients? “You know — the awkward ones.” Awkward how? “Jessica knows.”

Jessica knew. Jessica had been doing it for eleven years. Jessica looked at the client name and just knew — that this one always negotiates so you pad it, that one pays late so you take a deposit, this other one’s a friend of the owner so you don’t apply the minimum. None of it written anywhere. None of it in the CRM. All of it correct, and all of it load-bearing.

There was no quoting process. There was Jessica, and a spreadsheet that recorded what Jessica had decided.

You cannot automate that. Not because it’s hard — because it doesn’t exist in a form anything can execute. The four hours a week weren’t the process. They were Jessica thinking.

So we spent two weeks writing down what Jessica knew. That was the project. The automation itself took three days afterwards, and it worked — and it worked because for the first time in eleven years, somebody had asked Jessica how she does it.

Then Jessica retired last year, and they still have her judgment. That turned out to be worth vastly more than the four hours.

---

Why automation projects fail

Not the tools. The tools are fine. Zapier, Make, n8n — arguing about them is a distraction people use to avoid the real problem.

They fail because the process doesn’t exist.

Every business believes it has processes. What it usually has is a set of habits, several undocumented exceptions, and two or three people who know things nobody has ever asked them. That works — it works beautifully, actually — right up until you try to hand it to a machine, and the machine asks the questions nobody’s asked in a decade.

That’s the moment automation projects die. Everyone assumed the discovery phase was a formality. It wasn’t. It was the project.

They also fail because people automate the wrong thing first. They pick the task they hate. But you hate it because it’s fiddly and requires judgment, which makes it the hardest thing to automate and the most dangerous to get wrong. The right first target is usually something boring you barely notice — copying data between two systems, because it happens two hundred times a month and needs no thought whatsoever.

And they fail silently. A broken workflow doesn’t throw an error into anyone’s face. It just stops, and nobody notices for six weeks, and by then there are four hundred records that never synced. Every automation needs to be able to tell you it’s broken. Almost none do. I have a personal thing about silent failure — I lost a business to one.

---

The Automation Ladder

Climb in order. Skipping rungs is how you end up with an expensive, confident, wrong system.

Rung 1 — Move data. System A to System B. No decisions, no judgment. Boring, high-volume, immediately valuable, almost impossible to get wrong. Start here even if it feels too small. Especially if it feels too small.

Rung 2 — Notify. Tell the right person when something happens. Still no decisions. Removes the “did anyone follow up?” tax.

Rung 3 — Rule-based decisions. If this, then that — where the rules are genuinely explicit and you can write them all down. This is where Jessica lives, and where the writing-down happens. Most of the value in most businesses is on this rung.

Rung 4 — AI judgment, with a human deciding. The model reads, classifies, drafts, suggests. A person approves. This is the sweet spot right now for almost every small business — most of the speed, none of the exposure.

Rung 5 — AI agents that act. The system takes real action without asking. Genuinely powerful, genuinely risky, and appropriate far less often than the market suggests. Only for reversible actions, with hard limits and logging, once rungs 1–4 are solid.

Almost everyone wants to start at rung 5, because rung 5 is what’s on LinkedIn. Almost everyone should start at rung 1. The money is on rungs 1–4 and it always has been.

---

What it costs

From $1,495, scaling with process complexity.

That entry point covers a well-defined, single-process automation — the rung 1 or rung 2 work that most businesses should start with. Multi-system workflows, real decision logic, or anything on rungs 3 and 4 costs more, and I’ll scope it before starting rather than after.

The bigger variable isn’t the automation. It’s how much of the process already exists in writing. If it’s documented, we build. If it’s Jessica, we’re spending two weeks on discovery first, and that’s the honest quote.

The process documentation is available on its own, without any automation attached. Sometimes that’s the right purchase — you get a written, owned, transferable record of how your business actually runs, and you might discover you don’t need the automation at all. That’s a legitimate outcome and I’d rather sell it to you than build something you don’t need.

---

Let me be honest about agents

Agents that take autonomous action are the most oversold thing in this industry right now, and I say that as someone who builds them.

They work. When the process is well-defined, the actions are reversible, the limits are hard, and there’s logging, they’re genuinely transformative.

They fail badly when any of that isn’t true — and they fail confidently, which is the problem. A broken script stops. An agent keeps going, doing something plausible and wrong, at machine speed, until somebody notices. That’s a different category of risk and it deserves to be treated like one.

So my rules, which I’ll hold to even if you push:

• Reversible actions only, unless there’s a human approving. An agent can draft the email. It doesn’t send the invoice.
• Hard limits. Money, volume, scope. Enforced in the system, not in the prompt. A prompt is a request; a system limit is a wall.
• Everything logged. If you can’t reconstruct what it did and why, you don’t have a system, you have a rumor.
• It must be able to say it’s stuck. Nothing that guesses when it doesn’t know.
• Rungs 1–4 first. Always.

If that sounds conservative for someone selling AI automation — I got taken apart once by a system doing something confidently and invisibly on my behalf. It’s shaped how I build. I’d rather tell you that up front than after.

---

On tools

People want me to say n8n. Sometimes it’s n8n — it’s excellent, self-hostable, cheaper at scale, and better when data privacy is a real constraint. Sometimes it’s Zapier, because you need two connections and no maintenance burden. Sometimes it’s Make. Sometimes it’s forty lines of code and no platform at all.

The tool is the last decision, and the least important one. Choosing the platform before you understand the process is choosing a van before you know what you’re moving.

Anyone who tells you their preferred tool before asking what you do is selling you their preferred tool.

---

Ask Jessica first

There was no quoting process. There was Jessica, eleven years of undocumented judgment, and a spreadsheet recording her decisions.

Writing that down took two weeks. The automation took three days. And when Jessica retired, they still had what she knew — which turned out to be the actual asset, and the thing nobody had thought to ask for.

Automation is mostly just asking people how they do their job. Everything after that is plumbing.

Let’s find out what you actually do — $997 Diagnostic →

---

About Stevie

I automate things for a living and I’m more cautious about it than most people selling this, for a specific reason: I lost a business to a system that was doing something confidently, invisibly, and wrong, on my behalf, for weeks.

So I build things that tell you when they’re stuck. It’s not the exciting pitch. It’s the one I’d want.

sosSTEVIE — for when your digital presence is broken, invisible, or about to be. (754) 302-4631

---