Blog
Where do you start with AI in your business?
· 3 min read
You have seen AI work for others and you want to get started, but as soon as you look around your own business, deciding where to begin turns out to be harder than expected. Customer service, invoicing, quotes, scheduling: AI seems to have something to offer everywhere, and precisely because of that it never gets beyond good intentions. A familiar situation, and one you can solve with a few clear criteria.
What makes a good starting process
Not every process is a suitable first project. The processes best suited to AI share a few characteristics.
First: the process is repetitive and happens often. Compiling a report once a quarter yields little gain if you automate it. Answering the same kind of email or assessing the same kind of document hundreds of times a week does.
Second: both the input and the desired output can be described clearly. If you can explain to someone what comes in and exactly what needs to come out, an AI system can learn it. If the assessment is largely a matter of feel, experience or internal politics, it is not a good starting point.
Third: the data already exists, and sits in a system you can access. A process that first requires weeks of collecting and cleaning data is not a quick win, however promising the process may otherwise be.
Fourth: a mistake costs time to fix, not serious damage. Start with a process where an AI error means someone corrects the result, not with a process where an error directly affects a customer, an employee or a legal obligation.
Typical candidates in an SME: pre-sorting incoming email or requests, summarising long documents for a colleague, filling in standard forms based on existing data, and answering frequently asked customer questions before a person takes over.
Where you are better off not starting
Knowing where not to start is just as important as knowing where to start. Two categories call for extra caution.
The first is any process in which an AI system makes a decision about a person on its own, without anyone reviewing it. Think of automatically rejecting a job applicant or automatically changing a payment arrangement. That is not only risky for your reputation, it also falls under stricter European rules. The EU AI Act explicitly names systems for recruiting and selecting staff, and systems that assess people's creditworthiness, as examples of high-risk AI, precisely because the outcome affects someone's livelihood. Such applications are subject to heavier requirements for documentation, human oversight and risk management, before you even start building. Take a look at the overview of high-risk systems in the AI Act if you want to know whether your process falls under it.
The second category is any process whose outcome you cannot measure. If after three months you cannot say whether things have become faster, cheaper or better, you cannot adjust course either. Better to start small and measurable than big and uncertain.
How to recognise this in practice
A good example of a process that meets all the criteria is a manual email process overflowing with repetition. At MEA Beveiliging, the company processed all installation reports from its dealers manually by email: 50+ dealers and hundreds of vehicles a year, resulting in chaos and delays. That process was repetitive, had clear inputs and outputs, and an assessment without hidden politics. As a result, it could be fully automated in a dealer and certification platform: dealers now manage their own installations and certificates are generated instantly. Today, 50+ dealers and 768 vehicles run through the platform. Details and background can be found on our projects page.
That is exactly the pattern to look for in your own business: not the most prestigious process, but the process with the most repetition and the least ambiguity.
What does this mean for your business?
A few steps you can start with tomorrow.
Make a list of five processes that occur frequently and take up a lot of time. Think of anything involving email, forms or documents; that is usually the richest hunting ground.
Test each process against the four criteria: does it occur often, are the inputs and outputs clear, is the data available, and is a mistake cheap to fix? Cross out anything that fails on any of the four.
For what remains, check whether it involves a direct decision about a person, such as hiring, dismissal, credit or a similarly weighty decision. If so, set it aside and handle it with extra care, or do not start on it yet.
From what is left, choose the process with the most repetition, and agree in advance how you will measure in a few months whether it has succeeded: less time, fewer errors, or faster replies to customers. Without that yardstick, you will not know afterwards whether you have achieved anything.
Not sure which process in your business is the best starting point, or would you like to talk through how to set up such a first step carefully? Feel free to contact us.