Categories, not brand names

A list of product names goes out of date in months. A list of categories does not, and it is more useful anyway — because the question is never “should we buy this tool”, it is “do we have this problem”.

So here are the categories that genuinely help a small business, what each one is for, and how to tell whether you need it.

1. Document reading

What it does. Reads scanned invoices, receipts and forms, and turns them into structured data.

Why it is first. High volume, dull, and instantly verifiable — you can see whether the number is right. For most small businesses this is the fastest payback available.

Do you need it? Count the documents you type in each month. Above fifty, it pays.

Where to get it. Your accounting or ERP system probably already has it. Check before buying anything separate. This is the most commonly duplicated purchase.

2. Writing assistance

What it does. Drafts emails, descriptions, articles, social posts. Rewrites and shortens.

Why it helps. In a small company one person often writes everything. Removing the blank page is a real saving.

Do you need it? If somebody is spending hours a week writing routine text, yes.

The rule: read every word before it goes out.

3. Meeting notes

What it does. Records a call, transcribes it, produces a summary and action points.

Why it helps. In a small team the person running the meeting is also the person taking notes, and doing both badly.

Do you need it? If your meetings produce actions nobody wrote down, yes.

Check the consent rules where you operate — recording a call usually requires telling people.

4. Customer support assistance

What it does. Summarises a customer’s history before you reply, drafts responses, suggests knowledge base articles.

Why it helps. In a small team whoever is free answers the ticket, and they may not know the history.

Do you need it? If support has outgrown one person’s memory, yes.

Start behind the agent, not in front of the customer. Summaries and drafts first. Chatbots much later, if at all.

5. Design and images

What it does. Generates or edits images, removes backgrounds, resizes for different channels.

Why it helps. Small businesses rarely have a designer.

Do you need it? If you are paying for stock photos or going without images entirely.

Note on rights: generated images are usually fine to use commercially, but check the terms. And never take images from a Google search — those belong to somebody.

6. Data and spreadsheets

What it does. Explains formulas, cleans messy data, summarises a spreadsheet.

Why it helps. Most small businesses still run on spreadsheets somewhere, and the person maintaining them is rarely a spreadsheet expert.

Do you need it? If you have data you cannot get answers out of.

What most small businesses do not need yet

Being direct here saves money.

Prediction and forecasting tools. They need years of clean history. If you close forty deals a year or sell four large contracts, a forecast is a guess with a confident interface.

Lead scoring. Same problem. Thirty enquiries a month can be sorted by a person in five minutes.

Chatbots. With low website traffic, a contact form and a phone number work better.

Anything requiring an integration project. The maintenance cost outlasts the enthusiasm.

The mistake that costs most

Buying five tools that overlap.

A separate writing tool, a separate meeting tool, a separate support tool, a separate document reader — and then discovering your business system already does three of them.

Before buying anything, check what you already pay for. Your ERP, your email suite, your accounting package. AI features are increasingly included rather than sold separately, and the ones inside your business system have an advantage: they can see your actual data.

A tool inside your system beats a better tool outside it, most of the time, because context matters more than features.

How to choose

Five questions, per tool.

What specific task does it do? If the answer is “optimises your workflow”, it is a label, not a capability.

What does it cost per month, and per user? Small subscriptions accumulate quietly.

Where does our data go? Ask whether it is retained and whether it trains their models.

Do we already have this? The question that saves the most money.

How would we know if it were wrong? Every useful business AI has a review step.

A sensible plan

1. Fix your data first. Duplicate customers, wrong product costs, inconsistent units. Every tool works better on clean data and worse on bad data.

2. Check what your existing systems include. Often more than you expect.

3. Start with document reading. Highest volume, easiest to verify.

4. Measure it. Hours before, hours after.

5. Add one more thing. Not four.

6. Review subscriptions quarterly. Cancel what nobody uses.

What to be sceptical about

“AI-powered” as a selling point. Ask what it does.

Tools that need all your data uploaded. Ask why, and what happens to it.

Anything promising to run itself. Every useful business AI has a person checking.

Free tiers for confidential work. They often use your data for training.

The short version

Three categories pay for themselves at small scale: reading documents, drafting text, and capturing meetings.

Prediction and forecasting need data volumes most small businesses do not have, however good the demonstration looks.

Check what you already pay for before buying anything. AI features are increasingly bundled, and the ones inside your business system can see your actual data — which usually matters more than the feature list.