AI for SMEs: where do you start?

What works and what's still hype

State of play in 2026: AI is mature enough for specific tasks, but not for everything. A few examples where AI actually delivers now:

• Reading and summarising text (emails, reports, quotes) is practical and reliable.
• Generating text in a fixed structure (quotes, email confirmations, product descriptions) works well.
• Image recognition (handwritten notes, product photos, licence plates) is usable at a level in 2026 that was unthinkable in 2022.
• Speech to text (phone calls, meetings, voice memos) is mature and cheap.
• Smart chatbots answering customer questions based on your own documents are reliable enough to deploy.

Where AI doesn't really work or only partly works for SMEs:

• Making full decisions without a human in the loop. AI can suggest, but the owner or staff member still approves.
• Complex legal or medical assessments. The risk of errors is too high for production processes.
• Generic 'AI strategies'. Lots of sales talk, little concrete delivery. Start with one concrete process, not with a vision.

The three sweet spots for SMEs

1. Repetitive writing

Quotes, emails, product descriptions, job ads, customer confirmations. Work that has a fixed structure but is always slightly different. This is the clearest win for most companies.

2. Reading incoming information

Categorising emails, reading work tickets, processing invoices, summarising forms. Work currently done manually by someone that AI can handle well, provided there's a fallback for unclear cases.

3. First-line customer contact

A chatbot that answers the first questions and either helps the customer directly or routes them on. Fed properly with your documentation, something like this catches 60 to 80 percent of simple questions. That saves your team hours per week.

Four questions before you start

1. Which process eats up most of your time right now? Start there. Not with what's technically interesting.
2. Is that process stable? A process that changes every month isn't a good candidate to automate.
3. Can you describe it in rules? If you can explain it to your team, AI or automation can do it too.
4. What do you do with uncertain outcomes? AI makes mistakes. Build a fallback up front: a human checks, or the user confirms.

What AI isn't

AI isn't an independent employee. Not a replacement for strategic thinking. Not a replacement for craftsmanship. It's a tool that can take over repetitive work, provided you set it up and monitor it well. Companies that understand that get the most value out of it. Companies expecting a miracle end up disappointed.

Three first steps

1. Inventory. Spend an hour writing down which processes cost time every week without customers seeing them. That's your long list.
2. Pick one. The most painful. Or the easiest to build. Ideally both.
3. Build small. Don't reorganise your whole business. One automation, one month of building, one team that's going to use it. Move on once that works.

If that feels hard, take our inspiration quiz. In two minutes we suggest a top three based on your situation. No obligation.