Every business owner has now been told to "leverage AI" by someone who couldn't say where. Here's the where. In a small or mid-sized business, AI pays off in the repetitive middle: narrow, high-volume tasks with clear right answers, built into the software your team already uses. It disappoints almost everywhere else. The good news is that two or three well-chosen automations are usually worth more than the entire "AI strategy" genre.
Where it pays
- Inbound documents. Supplier invoices, purchase orders, delivery dockets — read, extracted, matched and entered into your systems, with a human approving rather than typing.
- Inbound email. Orders, queries and complaints classified, routed and drafted for reply, so the inbox stops being a job.
- Quoting. Draft quotes assembled from your own pricing rules and history, ready for a salesperson's judgement instead of their hour of assembly.
- Job notes and handovers. Field notes summarised into structured records; the 5pm data-entry ritual retired.
- Internal questions. Staff assistants answering from your procedures and product data — not the open internet — so the same question stops interrupting your best people.
The pattern in every winner: high volume, low variance, a clear right answer, and a human left in charge of judgement. Measure in hours saved per week; if you can't, it's not one of your two or three.
Where it disappoints
- The website chatbot bolted on because competitors have one. It answers what your FAQ already answered and frustrates everyone else.
- Anything customer-facing without review where a wrong answer costs real money or trust. Keep humans on send.
- Judgement calls — pricing exceptions, hiring, anything contractual. AI drafts; people decide.
- "AI transformation" bought as a strategy engagement. If it doesn't name specific tasks and hours saved, it's theatre.
The safety question, answered plainly
"Is our data safe?" — it can be, if the automation is engineered rather than improvised. Built properly, everything runs under your own accounts, your data stays in your systems, providers are chosen on their data-handling terms, and nothing of yours trains anyone's models. The real-world risk in most SMBs isn't the AI provider — it's staff quietly pasting customer data into free consumer tools because nobody built them a safe alternative. The fix for shadow AI is a sanctioned tool that's actually better.
How to start without a strategy deck
- Find the highest-volume repetitive task in the office — ask "what do you retype most?" and believe the answer.
- Automate that one task inside the existing workflow, human in charge.
- Measure the hours for a month. Keep it if it pays; kill it if it doesn't.
- Repeat. Stop when the next candidate wouldn't pay.
This is unglamorous, which is exactly why it works. The businesses quietly winning with AI aren't the ones with a transformation roadmap — they're the ones whose invoices enter themselves.