AI for business: where it helps, where it does not, and where to start

AI for business: what it means in practice

When someone says "AI" today, they usually mean a language model, like the one behind ChatGPT. It reads and writes text: it understands an email, summarises a document, suggests a reply. It does not think like an employee and knows nothing about your company beyond what you show it.

A small business can use it in three ways. First, ready-made tools such as ChatGPT, where someone on the team types and copies. Second, AI features built into software you already use. Third, a system built around your work and connected to your email, spreadsheets or ordering system. The higher on the list, the easier it is to start. The lower on the list, the more work the system takes over from a person.

Where it helps today

• Drafts Replies to enquiries, quotes, product descriptions, emails to suppliers. A person reviews and sends, but does not start from a blank page.
• Customer questions Opening hours, address, services, how to book. A chatbot answers from the information you have given it and hands over to a person anything it does not know.
• Data from documents Invoices, contracts, orders, filled-in forms. AI pulls the fields you need into a spreadsheet, and a person checks the doubtful ones.
• Sorting The inbox, enquiries and messages are split by topic and urgency before anyone opens them.
• Summaries A long email thread, meeting notes or a call recording, reduced to a few lines and a task list.
• Searching your own documents The team asks "what does the contract with this supplier say" and gets an answer with a reference to the document, instead of searching folder by folder.

Where it does not help, or not yet

• When the process is chaos If nobody knows who answers the enquiries, AI will only make the mess faster. Sort out the process first.
• When nobody checks A model can give a wrong answer with the same confidence as a right one. Where a mistake is expensive, a person has to come before the result.
• Exact figures Prices, totals, stock and tax are calculated by ordinary software, not by a language model. The model can prepare, but it should not have the last word on numbers.
• Customer relationships A complaint, a negotiation or bad news. Here the person is the reason the customer stays.
• When the task is rare Something that happens once a month is rarely worth building for. It is usually cheaper for someone to do it by hand.

How to choose the first task

A good first task has five signs. The more of them it has, the more likely it is to be worth it.

• It repeats often Every day or every week, not once a year.
• It takes the time of someone whose time is expensive Often that is the owner or the most experienced employee.
• The result is quick to check A person can see at a glance whether the draft or the spreadsheet is right.
• A mistake can be fixed If something goes wrong, it can be undone before it reaches a customer or a bank.
• The data exists The emails, documents or records are already somewhere in digital form, not only in someone's head.
• If you do not know where to start The free 5-minute check ranks the leaks in your business by what they cost, using your own numbers.

The risks

• Data What you send to the model and where it is processed. Customers' personal data falls under the data protection rules. Ask the provider whether the data is stored, where, and whether it is used for training.
• Mistakes AI sometimes invents facts that sound plausible. That is why everything that goes out passes through a person.
• Over-trust After a few weeks without a mistake, people stop checking. That is when a mistake slips through. Checking has to be part of the process, not a matter of mood.
• Dependence on a provider Models and their prices change. Ask what happens to your work if the provider changes its terms or stops the service.

What it costs to own

The cost does not end with the build. Before you start, be clear about who pays for and who is responsible for each of these.

• The build Connecting to your email, spreadsheets or systems, rules for when a person is asked, and a trial with real cases.
• The model fee Model providers usually charge by how much text is processed. Ask how it is calculated and what happens when there is more work.
• Maintenance Models get updated, your processes change, and connected software changes how it works. Someone has to make sure the system keeps working.
• People's time Someone reviews the drafts, approves and corrects. That time is part of the bill.

How we help

We build working software for small businesses, and AI is only one of the tools. Often the better answer is an ordinary automation without AI, and we say so. The €390 audit maps your processes, works out where the time and money go and ranks what is worth building first. If by your numbers the audit will not pay for itself, we tell you not to buy it.