How to identify the first AI use cases in a small business

Most small businesses don’t have an “AI gap” problem. They have a focus problem. AI could do hundreds of things, and that’s exactly what paralyses people. The useful question isn’t “what can AI do?” but “which specific task costs us hours every week and could be done just as well with help?”.

This guide is about finding those first use cases without launching a six-month project or filling the company with tools nobody will use.

The simple rule: repetitive, written, low-risk

A good first use case meets three conditions at once:

• Repetitive. It happens several times a week, not once a quarter. The savings compound.
• Text-based. Writing, summarising, rephrasing, classifying, pulling data out of a document. That’s where today’s AI is strong.
• Low-risk. If the first draft is imperfect, a person reviews it in seconds and nothing bad happens.

If a task ticks all three, it’s a candidate. If it misses one, leave it for later.

Inventory the week, not the company

Don’t try to map the whole operation. For one week, note the written tasks that repeat: answering the same kind of email, preparing similar quotes, drafting profiles, summarising meetings, turning rough notes into a clean document. Put an honest minute estimate next to each.

It’s almost always the same suspects:

• Customer communication: answers to frequent questions, follow-ups, confirmations.
• Quotes and proposals: starting from a previous one and adapting it.
• Internal documentation: turning what someone knows by heart into text others can follow.
• Summaries: of calls, long email threads, dense documents.

A concrete example: a solar panel installer spends hours every week replying to customers asking about timelines, subsidies and next steps after a quote. It’s repetitive, it’s text, and it’s low-risk if someone reviews before sending. A perfect use case to start with.

Before you automate, delete

Here’s the step almost nobody takes. Before applying AI to a task, ask whether that task should exist at all. I use a simple mental order, DAD + V: Delete, Automate, Delegate, Validate.

First delete: many reports and emails happen out of habit, not because anyone needs them. The best automation is the task you stop doing. Only once a task survives the “delete” test does it make sense to automate it (with or without AI), delegate it, or, if you keep doing it yourself, at least validate it with a checklist. Applying AI to a useless process just produces garbage faster.

Test one, measure it, then add the next

Pick one use case, the one that costs you the most time. For a week, do it with AI’s help and record two things: how long it took, and how much you had to correct. Compare with the “before”.

If you save real time and quality holds, turn it into a mini-procedure (a saved prompt, a context document) and move to the next. If you don’t save time, drop it without drama. This loop, one task at a time, avoids the most common mistake: buying ten tools, abandoning nine, and measuring none.

Context matters more than the prompt

When a first use case “doesn’t work”, it’s almost never the prompt. It’s that the AI doesn’t know your business: your prices, your tone, your terms, your typical mistakes. Gathering that into a document the AI can read (what I call a second brain for agents) changes the results more than any prompting trick. It’s also the first step toward later moving from chat to agents that do real work.

Where to go next

To see how this fits into hands-on training for your team, look at the AI training for small business page. And once your use cases are chosen, continue with how to train a small team in AI without losing weeks in theory.