Where do I start with AI in a small business with no technical staff?

Time three recurring processes for a week. Pick the one that's frequent, rule-based and stable. Build the simplest version — usually a no-code automation with one model step — and run it alongside the manual process for two weeks. Decide on day thirty with numbers.

Which tool should I use?

Decide in week three, not week one. For most first projects n8n, Make or Zapier with a model call is enough. If your one-page spec is too complex for that, it's time to talk to someone who builds.

What's a good first workflow?

Inbound enquiry response, intake paperwork, follow-up sequences, the weekly report, or scheduling that lives in one person's head. Whichever passes the three-question test with the most hours.

Can AI talk to my customers in month one?

It can draft; a person should approve. Unsupervised customer contact is rarely the first win and often the first embarrassment.

What if it doesn't work?

You still have a timed, documented process and a number. That's a better outcome than most companies get from a platform purchase, and it makes the next attempt cheaper.

AI for business

Where to start with AI in a small business: a 30-day plan

Author

AI Cubed

Published

September 9, 2026

Read time

9 min

Most advice on where to start with AI comes from someone selling the place to start. This plan doesn't. It's thirty days, one workflow, and a measurement at the end — the same shape as the first month of an engagement with us, minus us. If you get to day thirty with a number, you've done better than most companies that bought a platform first.

It assumes you have no technical staff. It does assume one person owns the month and can protect a few hours a week for it.

Week one: time the work

Pick three recurring processes — inbound enquiries, onboarding paperwork, the weekly report, scheduling, invoicing, whatever your team does most. Time each as it actually happens, including the parts that aren't in any procedure: the chasing, the retyping, the waiting for someone to be free. Write down every hand-off and every field entered twice. At the end of the week you'll have hours per week per process, and it will be more than you thought.

Don't pick a tool this week. Don't read about AI this week. Just measure.

Week two: pick one with the three-question test

The three-question test
Question	Pass	Fail — and what to do instead
Is it frequent?	Daily or weekly	Quarterly — leave it; automation pays back on volume
Is it mostly rules?	'If X then Y' with a few judgment calls a person can still make	Mostly judgment — a bad first project; consider it later with a human in the loop
Is it stable?	Same process for the last few months	Changes monthly — document and settle it first; automating a moving target wastes the build

Pick the process that passes all three with the most hours. Then write one page: the inputs (where the work arrives), the rules (what a person does with it), the output (what 'done' looks like), and the exceptions (when a person must decide). That page is the specification, whoever builds it.

Week three: build the simplest version

Now, and only now, a tool. For most small-business first projects the simplest version is a no-code automation platform — n8n, Make or Zapier — handling the rules, with a single model call handling the judgment step (classify this enquiry, extract these fields, draft this reply), connected to the tools you already use. If your one-page description is genuinely too complex for that, a small custom system may be right; that's the point at which to talk to an implementation partner.

Connect to the systems you already use. Don't move data into a new tool to make the automation easier.
Send a person every output for the first two weeks. The system proposes; the person approves.
Log every run: input, output, and whether the person changed it. That log is week four's data.

Week four: run in parallel and measure

Both the person and the system do the work. You're measuring three things: hours the person spent versus hours reviewing the system's output; errors made by each; and exceptions the system couldn't handle. Ask the person who used to do it what they think — they'll tell you where it's wrong before any dashboard does.

Day thirty: decide

If…	Then…
Hours saved are real and errors are equal or lower	Switch fully, keep the person reviewing exceptions, and pick the next workflow with the same test.
Hours saved are real but errors are higher	Keep the parallel run; fix the rules or the model step; decide again in two weeks.
Hours saved are small	Stop. You still have a timed, documented process, which is worth more than the automation was going to be.
The build was too hard for the tool	That's the signal to bring in someone who builds — with your one-page spec and your numbers in hand.

What to avoid in the first month

Buying a platform in week one.
Starting with the most painful process instead of the most rule-based.
Letting the system talk to customers without a person approving, in month one.
Any vendor whose answer to 'where do I start' is their product.
Skipping the measurement. If you can't say the number on day thirty, the month didn't count.

What AI Cubed does about this

Weeks one and two are our diagnostic; weeks three and four are our first build, done by our engineers with the measurement built in. If you'd rather not run the month yourself, see where to start with AI or AI automation for small business.

Frequently asked questions

Start here

See where your operation is losing time.

Twenty minutes with an operator, not a salesperson. We'll name the one bottleneck costing you the most — and tell you whether it's worth fixing with software at all.

Book your free 20-minute consult→

20 minutes · video call · no preparation needed