Most advice about AI for small business is written by people selling it. Here is a flatter view: a few things it does well enough to rely on, and a few it doesn’t.
The numbers, briefly
Adoption is real, not a fad. In 2025, 58% of US small businesses reported using generative AI, up from 40% a year earlier. Around 90% of owners who use it say it saves them time each week, and for administrative work the average saving lands near 6.8 hours a week — close to a full working day.
That is the headline. The detail is that the saving is concentrated in a handful of tasks, and spread thin everywhere else.
Where it earns its keep
Drafting the first version of something. Proposals, service descriptions, replies to common enquiries, job ads. AI is good at getting you to 70% of a draft in a minute so you spend your time editing rather than staring at a blank page. The edit is not optional — the draft is a starting point, not an output.
Turning messy input into structured data. Pasting an email thread and getting back a summary with action items. Reading 40 customer reviews and grouping the complaints. Pulling line items off a scanned invoice. This is tedious human work that AI does in seconds and rarely gets meaningfully wrong.
First-line support triage. Not replacing a person — sorting the queue. Tagging tickets by topic, drafting a suggested reply for a human to approve, flagging the angry ones. The human still sends the message.
Search across your own documents. “Which contract has the 30-day termination clause?” against a folder of PDFs. Set up once, useful for years.
Where it wastes your money
Anything where being confidently wrong is expensive. Legal, tax, medical, financial specifics. AI will give you a fluent, plausible, wrong answer with no signal that it’s wrong.
Fully automated customer-facing output. An unsupervised chatbot speaking for your brand, auto-published content, auto-sent emails. The failure mode is public and you don’t find out until a customer does.
“AI strategy” as a project. Buying a platform before you have a specific, boring task that is costing you hours. The businesses that get the 2.3× productivity difference are the ones that trained their people on a real workflow — not the ones with the biggest subscription.
How to start
Pick one task that eats a few hours every week, is low-stakes if it’s wrong, and produces something a person checks before it goes anywhere. Run it manually through a general tool for a month. If it holds up, that is the one worth wiring into your systems properly.
That is also where we come in: once a workflow has proven itself, we build it into your tools so it runs without someone copy-pasting between tabs. See workflow automation and AI infrastructure, or just get in touch and describe the task.
