AI automation for small business: a sane starting point

What is realistic at small-business scale

A ten-person business does not need an automation strategy. It needs two or three specific things to stop eating people alive. In practice that means picking the workflows that run every single day and taking the human out of the mechanical parts, while leaving the judgment where it is.

The realistic first year looks like this: one automation built and running properly within a month, a second one a month or two later once the first has proven itself, and a habit of asking whether new manual work should exist before someone is hired to do it. That is it. Businesses that try to do eight at once end up with eight half-working workflows and nobody who understands them.

The other thing that is realistic at this scale is doing it yourself. The modern tools are genuinely usable by non-developers, and the person who knows the process is usually the best person to automate it. Where owners get burned is not the building. It is choosing the wrong first workflow, or building something with no error handling that fails silently for six weeks.

The three tools you will hear about

Zapier is the easiest place to start. It connects thousands of mainstream apps, the interface assumes you are not a developer, and simple two-step or three-step automations take minutes. It gets expensive and awkward when workflows grow long, branch a lot, or need proper error handling, because pricing scales with how many tasks you run.

Make sits in the middle. You build scenarios visually on a canvas, which makes multi-step flows with branches much easier to see and debug than a linear list of steps. It is generally cheaper per operation than Zapier at volume, and the learning curve is a little steeper but not much.

n8n is source-available and self-hostable. You can run it on your own server, keep data on your own infrastructure, write code steps when the visual nodes will not do, and version-control your workflows. Its cloud plans price by workflow execution rather than by individual step, which suits long workflows. It leans developer, so it is the right pick when you have technical help and the wrong one when you do not.

You need one of these, not three. Pick on the shape of your workflows and the technical skill actually available to you, then stay there long enough to get good at it.

Build it yourself, buy something off the shelf, or hire someone

Buy off the shelf first. Before you automate anything, check whether the software you already pay for does it. Most CRMs send follow-up sequences. Most accounting packages chase invoices. Most job management tools schedule reminders. A surprising number of automation projects are really a request to turn on a feature the business is already paying for.

Build it yourself when the workflow crosses two or three systems and no single product covers it. Start with the smallest version, run it alongside the manual process for a fortnight, and only then switch off the manual step. Write down what the workflow does and where its credentials live, because in a year nobody will remember.

Hire someone when a failure costs real money, when the workflow touches five systems, or when the only person who understood it has left. Ask for a fixed price against a named list of workflows, documentation you can read, and accounts that stay in your name. An automation you cannot maintain or move is a liability dressed as an asset.

What it actually costs

Tools: the entry tiers of Zapier, Make and n8n cloud are cheap, and self-hosted n8n costs you a small server. A typical small business running a handful of automations is somewhere between nothing and a few hundred dollars a month in subscriptions. If you add AI steps that call a model, budget a bit more per month depending on volume, though for text drafting at small-business scale that is usually small.

Build help: freelancers commonly publish hourly rates, and agencies quote per workflow or per project, with ongoing support offered as a monthly retainer. Both are worth it for the right job. Neither is worth it for a workflow you have not scoped, because you will pay for the scoping either way and it is cheaper to do it before the meter is running.

AI Consulting is A$1,195 per month + GST, with a call each fortnight, clarity on what to build in Claude or ChatGPT and practical guidance on how to do it. No lock-in.

When advice pays for itself

It pays for itself when it stops one wrong build. If you are weighing up quotes that run into the thousands, an hour spent working out whether the workflow is worth automating at all is cheap insurance. It also pays when it turns a vague brief into a scoped one, because that is usually the difference between a fixed price and an open-ended retainer.

It does not pay for itself if you already know exactly what you want to build and it is one Zapier connection between two apps. Go and build that. Come back when the list gets long or the quotes get strange.

Custom development is A$150/hour + GST, with a separate project scope, estimated hours and terms before work starts.