Field note · Applied AI

How to Use AI in a Small Business Without Getting Burned

Published June 28, 2026 · Empire Publishing

Short answer: Put AI on the high-volume, low-stakes work — email, drafts, summaries, answering repeat questions — and keep one human glance on anything that reaches a customer, touches money, or can't be undone. That's the whole rule: use the leverage, keep it on a leash. Done that way, AI can give a typical owner back 10–20 hours a week without the disasters.

Where AI actually earns its keep

You don't need a strategy or a stack of 100 tools. You need AI pointed at the tasks that already eat your week:

• The admin avalanche — drafting and triaging email, paperwork, data entry, turning messy notes into clean documents.
• Your own files answering questions — instead of digging through folders, ask across your documents and get the answer with the source.
• Marketing like you have a team — social posts, email, images, even short video, on a tiny budget.
• Never dropping a customer — fast first replies, a simple chatbot done safely, a follow-up system that doesn't forget.

Each one is a recurring task with a clear input and output — exactly what AI is good at, and exactly where a guardrail is easy to add.

The five ways AI burns a business — and the one fix for each

1. It lies to a customer. A confidently wrong answer goes out under your name. Fix: nothing reaches a customer without your glance, and ground any chatbot in your real, approved info.
2. It leaks your data. Private details pasted into a public tool can stop being private. Fix: never paste customer data, passwords, or secrets into a tool you don't control — or run a private model for the sensitive work.
3. It produces slop. Generic AI text cheapens the brand you've built. Fix: use AI for the first draft, never the final voice — you edit, you ship.
4. It wastes money and time. Subscriptions and rabbit holes add up with nothing to show. Fix: tie every tool to a specific task and a specific hour saved; cut what doesn't earn it.
5. It makes an irreversible call. The real danger is AI acting, not just talking. Fix: never let AI move money, delete records, or send legally binding messages on its own. A human approves anything you can't take back.

Notice that all five fixes are the same shape: a human stays on the loop for the things that matter. That's the leash.

The leash, in one paragraph

You don't make AI safe by understanding how it works — you make it safe by deciding, in advance, what it's allowed to do without you. High-volume and low-stakes? Let it run. Customer-facing, money-touching, or irreversible? It drafts; you approve. That single boundary is what separates owners who get 15 hours back a week from owners who get a lawsuit, a leak, or a wasted quarter. The upside and the discipline are not in tension — the discipline is what lets you take the upside.

Frequently asked

How can a small business use AI safely?

Point it at high-volume, low-stakes tasks, and keep one human glance on anything customer-facing, money-touching, or irreversible. Use the leverage, keep the leash.

What are the biggest risks?

Wrong answers to customers, data leaks, generic "slop," wasted cost, and irreversible mistakes — each fixed by keeping a human on the loop for what matters.

Can a chatbot get me in legal trouble?

Yes — you can be held to what your chatbot promises. Ground it in approved info, limit what it can say, and route sensitive cases to a person.

← More field notes