What AI can actually do for a small business

Most of what's written about AI for business is either breathless or vague. What I've found building AI features for real products is narrower and more useful: AI is excellent at a specific class of tasks, mediocre at others, and the wins come from finding the first kind in your workflow.

Where it reliably saves hours

• Drafting. First drafts of emails, posts, descriptions, and replies in your voice — with a person reviewing before anything goes out.
• Reading. Turning documents, forms, and photos into structured data: an invoice into line items, a photo of a form into fields, a test strip into readings.
• Sorting. Tagging, categorizing, and routing incoming messages, requests, or records so the right thing lands in front of the right person.
• Answering. Assistants that respond from your own material — policies, product details, past tickets — rather than the whole internet.
• Summarizing. Long threads, call notes, and reports condensed to what matters.

The pattern: tasks a person does many times a day, where the judgment is light and the input is text or images.

Where it doesn't (yet)

• Decisions with real consequences and no human check. AI is a fast first pass, not a final authority.
• Anything that must be exactly right every time. It's confident and usually correct, which is not the same as always correct.
• Replacing a process nobody has defined. AI automates a workflow; it doesn't invent one.

What it looks like in practice

Two examples from my own work: a marketing engine for local businesses that writes ready-to-post promos every morning, tuned to the weather and each shop's slow days, and a water-chemistry tracker that reads a test strip from a phone photo. Neither is “an AI product.” They're ordinary tools where AI does the tedious middle step.

Keeping costs predictable

AI calls cost money per use, and an unguarded feature can run up a bill. Every AI feature I ship is metered and capped — per user, per tenant, per day — with the model chosen to fit the task rather than the biggest one available. Your data stays in your systems and isn't used to train anyone's models.

How to start

• List the repetitive screen tasks in your week. The ones that make you sigh are the candidates.
• Pick one with clear inputs and outputs — a document in, a record out.
• Add AI to the tool you already use, rather than adopting a new one for its own sake.
• Keep a person in the loop until the results have earned trust.

If you'd like a second opinion on where AI fits in your business — including an honest “it doesn't, here” — discovery is free.