Somewhere in your company, someone is retyping data every week. From an email into a spreadsheet. From a PDF into the accounting package. From a form into a system that doesn't talk to the one next to it. Nobody enjoys it, everybody does it anyway, and somehow it never makes it onto the agenda.

That's where AI automation for small and medium businesses begins. Not with a strategy, not with a platform, but with one concrete task you want off your list.

AI automation for SMEs begins with one task

Most stories about AI in business are about transformation. That's for companies with a thousand employees and an innovation budget. When you're five to fifty people, it works differently: you don't have a department you can free up, you have a team that's already busy.

The good news is that you don't need one. Today's AI models can do things that five years ago were a months-long custom project: understand text, pull data out of messy documents, classify something, draft a piece of writing. What you still need is someone to tie that model neatly into your process. That's usually a few weeks of work, not a year.

How do you spot a task worth automating?

Walk through your week and look for tasks that meet these four points:

It repeats. Daily or weekly, not twice a year. Something you rarely do is rarely worth building.

You can explain the rules. If you can teach a new colleague how to do it in fifteen minutes, you can teach a model too. If you can't, there's craft or responsibility involved and it belongs with a human.

The input is already digital. Email, PDF, form, export. Paper that has to be scanned first makes the project longer.

A mistake is recoverable. A miscategorised purchase invoice you fix. A wrongly sent payment you don't. Start on the side where mistakes are cheap.

If a task meets all four, you have a candidate. If it meets three, it's often still worth an afternoon's look.

Four examples we see often in SMEs

Pulling requests out of the inbox. Customers send their question as free text. Someone reads along, extracts the name, product, quantity, and desired date, and puts it into the system. A model handles that first pass just fine, with a small screen next to it where your employee confirms or corrects in two seconds.

Reading purchase invoices. Every supplier has its own layout, so fixed templates only half work. A language model reads the messy ones too and hands you supplier, invoice number, amount, and VAT.

Sorting incoming questions. What's an outage, what's a quote request, what's a billing question? Automatically labelling and routing them saves no thinking, but it does save the daily triage.

Recurring reports. Pull figures from two sources, combine them, add in plain language what stands out, and land it in the inbox every Monday morning.

Each one small. That's exactly why they work.

Work out for yourself whether it pays off

Take the task you have in mind and fill in: how many times per week, how many minutes each time, at what hourly rate.

Twenty times a week, ten minutes each, that's a good three hours. Around a hundred and fifty hours a year. A solution is allowed to cost something against that.

Add two things you don't express in hours. The mistakes that disappear once something is no longer retyped by hand. And the fact that the work simply carries on when the person who always did it is on holiday.

Subtract one thing: automation is never a hundred percent. Count on a part that a human keeps reviewing. A solution that takes eighty percent of the work off your plate and lays out the rest neatly is, in practice, better than one that claims everything and quietly gets something wrong now and then.

What does it cost to have this built?

At CodeWerf the prices are simply on the site. Four packages, each with a fixed amount and a fixed scope, so you know up front what you're getting into. No three-week quoting process just to find out whether it fits your budget at all.

For most automations like the ones above, you're in the bottom two packages. Have a look at the pricing and see which package fits your task.

Have you already built something with ChatGPT or Claude?

More and more often, business owners come to us who've already tried it themselves. There's a prompt, or a little script, or a loose tool that works in their own browser. It does what it's supposed to, but it isn't a company asset yet: there's no login, colleagues can't reach it, there's no server, there's no backup, and nobody knows what happens when it breaks.

That's not a failure, it's the best possible starting point. You've already done the hardest part, namely proving the idea works. What's left is taking it from prototype to production: a proper interface, users and permissions, hosting, logging, and someone who calls when it breaks.

Name one task, and we'll take a look

You don't need an AI strategy. You need to name one task you'd rather be rid of.

Already know which one? Send it to us and within a day we'll tell you whether it can be automated, which package it fits, and what it saves you per year. If it turns out it can't pay off, we'll tell you that too.