AI tools for small businesses

Why it matters

For a business with fewer than twenty staff, the owner and a few key people do everything, and time is the scarcest resource. AI tools that shave an hour a day off admin, customer replies and bookkeeping are worth more to that business than almost any other technology investment, because the hours go straight back into selling and delivering. The flip side is that wasted subscriptions, half-adopted tools and a privacy incident cost the same small business proportionally more than they would a large one. Choosing well matters.

Start with the job, not the tool

The most reliable way to pick AI tools is to begin with a week of honest time tracking. Where do the hours go that do not directly earn revenue? For most small businesses the answer is some mix of writing and replying, chasing and coding transactions, scheduling, and producing documents from information that already exists somewhere else. Those are exactly the tasks generative and recognition models handle well.

Rank those tasks by hours consumed and by how costly an error would be. The ideal first candidate is high in hours and low in error cost: internal meeting notes, first drafts of routine emails, receipt capture. Leave anything customer-critical or compliance-heavy until you have a few months of experience with how the tools behave.

• Track a typical week: what you do, for how long, and whether it earns revenue.
• Shortlist tasks that are repetitive, text-heavy or data-entry-shaped.
• Rank by hours saved versus the cost of a mistake.
• Pick one. Resist the urge to fix everything at once.

Three ways to get AI: built in, bought, or built

Built-in AI is the features your existing software already offers: your accounting platform coding transactions and chasing invoices, your email suggesting replies, your CRM summarising calls, your website builder drafting copy. These are usually the best place to start because the data is already there, the vendor has handled privacy and integration, and the cost is often included or a modest add-on.

Bought tools are standalone AI products: general assistants, meeting transcribers, image generators, document-extraction services, customer-service chat tools. They are more capable in their niche than built-in features, but each adds a login, a data flow and a bill. Built solutions use a model provider directly, with a developer connecting it to your data and processes. This is the most powerful option and the most expensive to create and maintain; it makes sense when a process is central to your business and no product fits, and usually only once you have outgrown the first two.

Questions to ask any AI vendor

Small-business software buying is usually a credit card and a free trial, which is fine for a to-do app and risky for a tool that will see your customer list. Before any AI product touches business data, get clear answers to a short set of questions. Where is data stored and processed, and does it leave Australia? Is your data used to train the vendor’s models? Can you export and delete it? What happens when the model gets something wrong, and who is responsible? How is the price calculated: per user, per request, per document? And what does the tool do when it is unsure: guess, or tell you?

Vendors who answer plainly are usually the ones worth trusting. Vague answers about "enterprise-grade security" without specifics are a signal to keep looking. Record the answers; you will want them when a customer or regulator asks how their information is handled.

Run a pilot that proves something

A pilot is a month-long test with a clear question: does this tool save enough time or reduce enough errors on this task to justify its cost and risk? Record the baseline first, how long the task takes today and how often mistakes happen. Give the tool to two or three people, not everyone. Agree what success looks like in numbers before you start. At the end, decide: expand, adjust, or stop.

Most pilots fail for predictable reasons: no baseline, so nobody can tell whether anything improved; too many tools at once; or a tool trialled on the hardest task rather than a representative one. One tool, one task, one month and one measure is enough to learn what you need.

How to choose and pilot an AI tool

1. Track a week of time Find the repetitive, text-heavy or data-entry tasks that eat non-billable hours.
2. Pick one task with low error cost High hours, low stakes: meeting notes, routine replies, receipt capture.
3. Check built-in features first Your accounting, email, CRM and website tools may already do the job.
4. Shortlist two or three products if needed Ask the vendor questions on data, training, export, pricing and errors.
5. Record the baseline Time per task and error rate today, before anything changes.
6. Run a one-month pilot with a few people Agree the success number in advance.
7. Decide: expand, adjust or stop Add the tool to your register and training if you keep it.

Real-world example

A Cairns tour operator picks its first AI tools

Reef Edge Tours runs snorkelling and island day trips with eight seasonal staff. The owner, Ana, tracks a week and finds the office team spends hours a day answering repetitive enquiry emails and reconciling booking payments against the bank. She discovers her booking platform has added an AI reply assistant that drafts answers from her own FAQ page, and her accounting software already matches most payments automatically if she switches the feature on. Both are included in plans she already pays for.

For the remaining gap, meeting notes and supplier correspondence, she pilots one general assistant on a business plan for herself and the office manager. The success measure is thirty minutes a day saved per person. After a month, enquiry replies are faster and more consistent, reconciliation is largely automatic, and the assistant has earned its small subscription. Two other products she was pitched, an AI social-media scheduler and a "revenue intelligence" dashboard, go on a list to revisit next season once the basics are bedded in.

Common mistakes

• Buying tools before defining the task A tool looking for a problem becomes an unused subscription. Start with where the hours go.
• Ignoring features you already pay for Most accounting, email and CRM platforms now include AI. Switch those on before buying anything new.
• Trialling on the hardest task A pilot on a complex, high-stakes process fails and sours the team. Prove value on something simple first.
• No baseline Without before-and-after numbers you cannot tell whether a tool helped, so decisions become opinions.
• Nobody owns the tool register Logins, data flows and bills multiply. Name one person to keep the list current.

KPIs to track

• Hours returned per week Baseline time minus time with the tool, across the team. The number that justifies the spend.
• Adoption rate Share of intended users actually using the tool weekly. Below half after two months means training or fit is the problem.
• Cost per hour saved Monthly tool cost divided by hours saved. Compare with what those hours are worth to the business.
• Tools on the register Every AI tool in use, with owner, data and renewal recorded. The target is all of them.

Frequently asked questions

How much should a small business spend on AI tools?

Often very little to start. Built-in features plus one general assistant on a business plan is typically a modest monthly cost per user. Spend more only when a pilot has shown a specific return.

Should I pick one vendor’s ecosystem or mix tools?

Lean towards the tools you already run your business on; their AI features work with data you already hold. Add specialist products only where the gap is real and the integration is manageable.

What about tools built specifically for my industry?

Industry tools can be excellent because they understand your documents and workflows. Apply the same vendor questions, and check that the industry knowledge is real rather than a general model with a new name.

When does a custom build make sense?

When a process is central to how you make money, no product handles it well, and you have someone who will own the solution long term. SMBH Consulting typically recommends exhausting built-in and standalone options first; custom AI applications earn their cost on core processes, not convenience.

Key takeaways

• Start with where your hours go, not with the tool catalogue.
• Use the AI already inside the software you pay for before buying anything new.
• Ask every vendor where data goes, whether it trains models, and how errors are handled.
• Pilot one tool on one task for one month against a recorded baseline.
• Keep a register of every AI tool, its data, owner and cost.