AI for business: what actually works today and what's still hype

Every week there's a new headline about AI changing everything. But if you run a real company with real employees and real deadlines, you need a clearer answer: what can AI actually do for me today?

The honest answer: some things very well, some things not at all, and a lot of things somewhere in between.

What works right now

Reading and sorting documents. AI can read invoices, contracts, forms, and emails, extract the relevant data, and sort it into the right place. If your team spends hours processing paperwork, this is the fastest win. Accuracy is above 95% for structured documents, and the system flags anything it isn't sure about.

Answering routine questions. If your clients or employees ask the same 20 questions repeatedly, AI can handle them. Not with a clumsy chatbot from five years ago, but with one that actually understands context and gives useful answers based on your own data.

Spotting patterns in data. AI is good at finding things humans miss in large datasets. Which clients are likely to leave. Which orders are likely to be late. Which costs are growing faster than they should. If you have data, AI can probably find something useful in it.

What doesn't work yet

Replacing judgment. AI can suggest, but it can't decide. Strategic decisions, client relationships, creative problem-solving: these still need humans. Anyone selling you "autonomous AI that runs your business" is selling a fantasy.

Working with messy data. AI needs clean inputs. If your data is scattered across emails, sticky notes, and someone's memory, AI can't magically organize it. You need to get your basics in order first. Digitalization comes before AI.

Small datasets. If you have 50 clients and 200 transactions, AI won't find meaningful patterns. You need volume. For most small businesses, the document processing and question-answering use cases are more realistic than the data analysis ones.

How to think about it

Don't start with AI. Start with the problem. What's slow? What's repetitive? What's error-prone?

If the answer to that problem involves reading, sorting, matching, or responding to predictable inputs, AI can probably help. If it involves nuance, relationships, or creative thinking, it probably can't. Not yet.

The companies getting real value from AI aren't the ones chasing the latest model. They're the ones who found one boring, repetitive task and let AI handle it. The savings are quiet but they're real.