AI in your business: where do you start?

Almost every business wants to do something with AI, but few know where to begin. A chatbot subscription is signed up for in minutes. An application that genuinely supports your processes takes more thought. This article explains how to choose a first step that is small enough to keep under control and large enough to make a difference.

Where do you start with AI in a business?

Start with a concrete task, not with the technology. Look for work that recurs often, involves a lot of text or documents and is currently done by hand. Think of sorting incoming email, copying data from invoices or answering customer questions with information that already exists somewhere in your organisation. Tasks like these have three properties that make AI work well: there is enough example material, the outcome can be checked and a mistake can be corrected. Avoid choosing, as your first task, one where a wrong decision has immediate serious consequences, such as automatically rejecting an application. Start with a defined group of users as well, so you learn quickly what works. The goal of a first application is not only to save time but also to gain experience: how do employees use it, where does it go wrong and which agreements about data are needed?

Which AI applications work well for businesses?

The applications that work best in practice are usually not the most eye-catching. They are tasks in which a language model reads text, understands it and turns it into something useful, within a process that already exists. A knowledge assistant answers questions based on your own manuals and work instructions and cites its source, so an employee can check the answer. Document processing extracts data from PDFs, scans and emails and prepares it in the right system. Classification determines what an incoming message is about and which department should pick it up. An AI agent goes one step further and carries out a defined action, such as creating a draft order. What these applications have in common is that they take over the preparatory work, while a person keeps the decision where that is needed. Common examples are:

• Answering customer questions with information from your own knowledge base
• Reading purchase invoices and preparing them for checking and posting
• Sorting incoming email and routing it to the right department
• Summarising long documents or meeting notes
• Drafting product descriptions from technical data

What data do you need to start with AI?

Less than you might think, but it has to be the right data. For most business applications you do not need to train your own model. An existing language model can already do a lot, as long as it has access to the information relevant to the task. For a knowledge assistant, up-to-date and reliable documents matter more than a large volume of data. Outdated manuals or contradictory work instructions produce outdated or contradictory answers. For document processing it helps to have a set of examples, so you can test how well recognition works before going live. Also look at access: which employee is allowed to see which information? A good AI application respects those permissions and never shows users anything they would not have been allowed to open without AI. Cleaning up and organising your sources is therefore often the first step, and a useful one in its own right.

How do you keep control of privacy and quality?

By making agreements in advance and building them into the application itself. Decide which data the AI may and may not use, for example no medical or financial personal data if that is not needed. Choose a provider with business terms that does not train on your data, and check where processing takes place. For organisations in the Netherlands, processing within the EU is the obvious choice. Also decide when a person needs to review the output. That matters less for a summary than for an action in your ERP. Log every outcome, so you can see afterwards what the AI did and on the basis of which information. Finally, assess quality with a fixed set of test questions or test documents. That way you can see whether a change in the model or in your sources makes the results better or worse, instead of relying on gut feeling.

Not sure which task in your organisation is the best starting point? In a free introductory call we look at your processes together and discuss where AI makes sense, and where it does not.

Last updated: September 23, 2026