AI for small businesses: where to start (and what to skip)

If you run a small business, where would you actually start with AI?

I’ve been asked this a lot lately. Usually right after someone has opened seventeen browser tabs, bought a trial of something with “AI” in the name, and still feels none the wiser.

Here’s the short version: don’t start with a tool tour. And, don’t start with an “AI strategy.” Start with one job that already hurts, something repetitive, expensive, or slow — then run a small experiment you can kill without embarrassment.

If that sounds obvious, good. Most of the industry is still selling the opposite.

The pressure to “do AI”

There’s a lot of noise designed to make SME leaders feel late. You’re not late. You’re early for your business because your business has its own processes, messy data, customers, and constraints. Copying someone else’s approach won’t fix that.

At Cariboo we partner with non-technical leaders. The pattern is familiar: pressure to adopt something shiny, a flurry of tools, then nobody can point to a single job that got faster, cheaper, or less annoying.

That’s not a technology failure, it’s a starting point failure. We’re still confusing solutions with strategies. We need to stop starting with solutions.

Start with the problem, not the platform

No business should invest in AI,  or any capability, without being able to answer a few plain questions:

1. What problem are you trying to solve?
2. Who feels it, and why?
3. If it was fixed, would it matter?
4. By how much : time, money, risk, reputation?

If you can’t answer those, you’re not ready for a tool. You’re ready for a conversation about the work.

In practice, that conversation usually lands on one of three places:

• Where hours vanish : admin, rewriting the same emails, chasing status, fighting spreadsheets.
• Where mistakes cost money : missed follow-ups, inconsistent quotes, handoffs that quietly drop.
• Where customers wait : slow replies, unclear next steps, support queues that never shrink.

Pick the one that hurts most this month. Not “AI for the business.” One job.

What a sensible first experiment looks like

These tend to teach you something without setting the building on fire.

Admin drafting works well: first drafts of emails, proposals, standard operating procedures, meeting notes with a human editing before anything goes out. FAQ and help drafts grounded in your real policies, reviewed by whoever owns those policies. Search over your own docs, so people stop rummaging through Google Drive for the same answer. Inbox or lead triage that sorts and summarises, while people still decide what gets a reply.

See the theme? AI does the boring middle whilst you keep ownership of the outcome.

What I’d leave alone on day one is equally important. Unsupervised customer promises. Financial, legal, or compliance decisions with nobody checking the output. Boiling the ocean with a giant knowledge base before you’ve proven one workflow. Automating a messy process as-is — you’ll just automate the mess, faster.

If the job needs judgement, trust, or an irreversible action, keep a human in the loop. That’s not Luddism. That’s how you stay in business.

Thirty days, not a transformation programme

Here’s a shape that works without becoming theatre.

In the first week, pick one job. Write the steps as they are today, not as you wish they were. Name an owner and agree what “done” looks like.

Next, baseline it. Measure time or error rate for a couple of weeks of the current way. If there's no baseline, there' s no proof — only vibes.

Then pilot. Use AI on the boring middle only. Same owner. Same definition of done. Write down what broke.

Finally: keep, fix, or kill. Compare to the baseline. If it isn’t clearly better, stop. If it is, write the playbook and expand carefully.

Thirty days is plenty enough to learn.

The awkward bits nobody puts on the landing page

Models can sound confident and still be wrong. Treat outputs as drafts. If your files and processes are chaotic, AI will amplify that, not magically tidy it. Confidence is not correctness, don’t ship unsupervised answers to customers. And know what tools can see, and what they keep.

One of our principles at Cariboo is transparency. If you can’t explain a pilot to your own team in plain English, you’re not ready to scale it.

The annoying reality

Most businesses don’t have an “I need an AI platform” problem. They have problems like: customer support takes too long; the sales team spends hours on manual data entry; onboarding is inconsistent; report generation eats time.

Those might benefit from AI. They might also benefit from better process documentation, clearer ownership, or hiring another person. Not every problem is a technology problem and not every technology problem needs a chatbot.

Tools aren’t strategies. Understanding the job is.

If you want a hand

Once you’ve got a real job in mind, you don’t have to invent the tooling from scratch. We built Ai Builder so you can explore and test purpose-built agents at your own pace, without needing a full engineering team. And if you’d rather have someone in the room, we do consultancy the same way we always have: discovery, co-design, and a safer path from experiment to something that actually ships.

Either way, we start with the problem, not the platform.

A few questions people ask next

Where should a small business start with AI?
With one painful, repeatable job. Map it, baseline it, pilot the boring middle with a named owner, then keep, fix, or kill.

Do I need an AI consultant?
Not always. If you can name the job, own the pilot, and review outputs, self-serve exploration can work. Bring someone in when the stakes are higher, the data’s a mess, or you’ve stalled without a clear next experiment.

What’s the cheapest safe experiment?
Two to four weeks on one admin or triage workflow, with a baseline, an owner, and a hard kill rule. No multi-year platform commitment required.

Will AI replace my team?
That’s not how we work. For most SMEs the useful pattern is augmentation: AI takes repetitive draft work; people keep judgement, relationships, and accountability.

So — what’s the job?

Bring one process you’re already tired of. We’ll help you decide whether it’s a good first experiment, and what “good” would look like in thirty days.

Book a free 30-minute consultation — no obligation, focused on what you’re trying to achieve.