What AI really means for your SMB (without the hype)

In the news AI seems on the verge of taking over every job. In practice I see something very different at SMB companies: colleagues still manually retyping the same invoice three times a week across four different systems. That is where the wins are.

What already works well

Three types of tasks where AI can be deployed within a week:

1. Answering questions based on your own documents. An AI assistant that knows your manual, warranty terms, technical documentation and old quotes answers 80% of customer questions immediately. Colleagues get time back for the harder cases. Example: at an installation company I supported, customers went from twenty-minute hold times to a direct chat answer. Not magic, but measurable.

2. Text and data processing. Categorising incoming emails, extracting invoice lines from PDFs, summarising customer calls. Work that always got done "on the side" now happens in the background.

3. First drafts of quotes and emails. Not to replace a human, but to get past the blank page. You keep control, AI types the first 70% for you.

Where it goes wrong

Companies deploying AI without a clear goal often get stuck on three points:

• Data is scattered. Documents in OneDrive, customer details in an old database, emails in Outlook. AI can only work with what it can find. Tidy up first, then automate.
• Nobody knows what the bot can and cannot do. Which information do you share, which not? Can the assistant mention prices? Who sees which conversations? This needs to be clear up front.
• Built without the people who use it. Only when the colleagues on the work floor co-build does the solution actually fit the work.

What it costs

A first AI assistant for customer queries is a two-to-four-week project, depending on how many documents need to be unlocked. From around five thousand euros you have a working system you can adjust and extend yourself. Monthly API costs sit between ten and a hundred euros for an SMB setup. Not nothing, but no barrier.

When a spreadsheet is still better

Honestly: not every process needs AI. If a task happens less than five times a week, a well-organised Excel or a manual step usually stays the most practical solution. AI only earns its keep when work is frequent and repeatable.

What you can do now

Don't start with "we're going to do AI". Start with the question: which task in the office makes colleagues sigh? That is your first case. The rest follows.

Also read

• RPA or AI? An honest choice for your process - when do software robots fit better than AI?
• How half a second of load time costs you 7% revenue - concrete steps to make your website faster

Want to spar on this freely? Schedule a coffee chat of thirty minutes. I listen first, advise next. No sales pitch.