# AI automation: decide before you build

> Five checks a team can run on one workflow before paying anyone to automate it.

Author: Alex Cinovoj, Founder & CTO, TechTide AI · CTO & Co-Founder, CPG Playbook
Published: October 18, 2026 (2026-10-18)
Reading time: 6 min read
Section: Field notes
Canonical: https://alexcinovoj.com/blog/ai-automation-for-small-business

Most owners asking about AI automation are not short on tools. They are short on a clear answer to one question: is this particular workflow worth automating, and can we trust it once it runs without us watching?

That question is cheaper to answer before anything is built. Here are the checks I would run on one workflow first.

## Five checks before you automate

### One: can you describe it in one paragraph?

If the person who does the work today cannot write down the steps, the inputs and what a good result looks like, an automation will copy the confusion faster. Write it down first.

### Two: can you check the result?

You need a way to tell a right answer from a wrong one without redoing the work. If you cannot, you will not notice when it quietly gets worse.

### Three: what does a mistake cost?

A wrong internal tag is cheap. A wrong invoice, quote or customer reply is not. The more a mistake costs, the more you need a person reviewing results before they leave the building.

### Four: who owns it after launch?

Name one person with the time and access to keep it running. Without that, small automations break on the first expired login and stay broken. More on this in [why AI pilots die at handoff](/blog/why-ai-pilots-die-at-handoff) .

### Five: what does it cost to run?

Count model usage, tool subscriptions and the hours someone spends maintaining it. Then compare that with the time it saves. Realistic numbers are in [what AI workflow automation really costs](/blog/what-ai-workflow-automation-costs) .

> **The short version**
>
> One workflow, a result you can check, a mistake you can afford, a named owner, and a running cost lower than the time saved. Missing any of these is a reason to fix the workflow before automating it.

## Start smaller than you think

1. Pick the single most repetitive task, not the most impressive one.
2. Run it with a person approving every result for two weeks.
3. Keep a list of every correction. That list is your test set.
4. Only remove the approval step once corrections have become rare and cheap.

## When to get a second opinion

- The workflow touches customer data, payments or anything regulated.
- It already works in a demo but nobody is sure it is safe to rely on.
- A vendor is proposing a build and you cannot judge the quote.

That is the decision I run. The [Systems Decision Audit](https://techtideai.io/audit) on TechTide AI reviews one workflow and gives you a written verdict: ship, fix, redesign or stop. Implementation, if it is needed, is scoped separately.

## Frequently asked

### Do you need an AI automation consultant?

Not always. If one person can describe the workflow, its inputs and what a bad result costs, you can often test a simple automation yourself. Bring in outside help when the workflow touches customers, money or private data and nobody on the team can judge whether it is safe to rely on.

### Which workflow should you automate first?

Pick one repetitive task with clear inputs, a result you can check, and a mistake that is cheap to catch. Avoid starting with anything where a wrong answer reaches a customer before a person sees it.

### What should I have ready before talking to anyone about AI automation?

A written description of the workflow, a handful of real examples, how often it runs, who owns it today, and what happens when it goes wrong. That is enough to decide whether to build, fix, or stop.

### How does Alex help with this?

Through the Systems Decision Audit on TechTide AI: one workflow reviewed, one written verdict on whether to ship, fix, redesign or stop. Implementation, if needed, is scoped separately.