It's Monday morning and the finance manager wants to know which customers brought in the most revenue last quarter. Revenue sits in the accounting package, orders in a second system and complaints in a customer service spreadsheet. Three sources, three slightly different answers. By lunchtime a colleague is still retyping figures. It's exactly the kind of situation that makes you wonder whether AI could help.
Data analysis with AI means software finds patterns, anomalies and forecasts in your data that you would otherwise hunt for by hand. It only pays off if your data is reliable and in one place, and you have a concrete question to answer. So start with one process, one question and a measurable goal, not with a tool.
What AI data analysis delivers for your business
The gain is rarely a spectacular insight. Usually it comes from work that currently takes too many hours or gets noticed too late:
- Time saved on recurring work: sorting and reading invoices, forms and emails happens automatically, and your team checks the exceptions.
- Spotting errors sooner: anomaly detection flags an unusual amount, a strange stock movement or a sensor drifting out of line before it becomes a problem.
- Better forecasts: predictions based on historical data help with purchasing, planning and capacity.
- Fewer arguments about numbers: when all sources come together, the same question gets the same answer everywhere.
You're not the only one getting started. According to Statistics Netherlands (CBS), 13.8 percent of Dutch micro-businesses used one or more of the seven AI technologies surveyed in 2025 (provisional figures). That says nothing about what it delivered, but it shows this is no longer an exotic experiment.
Which techniques fall under data analysis with AI
The term is broad. In practice, these are the applications you'll run into most:
- Descriptive and diagnostic: what happened and why? Dashboards and reports fall under this, with AI helping to find connections.
- Predictive: what is likely to happen, such as demand, churn or payment behavior.
- Classification: automatically sorting incoming items, such as tickets, documents or requests.
- Anomaly detection: finding deviations in large volumes of data that a person would overlook.
- Document processing and routing: extracting data from documents and getting the right person or process to pick it up.
You don't need to do everything at once. Choose the technique that fits your question. If you want to know what went wrong last month, a good dashboard is often enough. AI only really comes into play when you want to look ahead or sort large volumes.
Get your data in order first
Many organizations have more data than they think, but it's scattered across separate systems, spreadsheets and databases. One sign you have work to do here first: the same question gives a different answer depending on which system you check. An AI model running on that foundation only makes the confusion faster.
A data platform brings the sources together in one reliable place. It doesn't have to be big: a data warehouse, a data lake or something simpler, depending on where you are now and where you want to go. As part of our Data and AI solutions, we design architectures like this so they stay scalable and maintainable.
Keep an eye on three things in this phase:
- Ownership: who is responsible for which source, and who is allowed to change it?
- Definitions: what counts as a customer, an order, revenue? Write it down before a model starts calculating with it.
- Connections: preferably pull data through APIs or fixed integrations, not manual exports.
How to start in five steps
A workable start is small and clearly defined. This is the order we recommend:
- Pick one question with an owner. For example: which requests sit around for too long? You can read how to find the right candidates in finding AI use cases in your own processes.
- Map your sources. Which systems and spreadsheets hold the data you need, and how up to date are they?
- Make the data usable. Remove duplicates, standardize and fill in gaps. This often takes more time than the model itself, and that's normal.
- Build a small first version. Run it alongside your current work and compare the output with what your team comes up with on its own.
- Agree in advance when you'll continue or stop. What counts as success, and when do you set it aside? How to set that up is covered in setting up an AI pilot: scope, success and stop criteria.
If you're working with a partner, a short intro call helps sharpen step 1. At Isatis you start with a 15-minute conversation in which you tell us what you want to achieve, and we tell you whether we can help. If it's not a fit, you'll hear that too.
What you do yourself and what you have built
Not every question calls for a custom solution. For a one-off analysis of a spreadsheet, a general-purpose AI assistant is often enough, as long as you spot-check the output and don't share sensitive data without clear agreements.
Building pays off when:
- The work recurs: the same analysis or sorting every day or week.
- The outcome matters: a mistake costs money, time or trust.
- It has to fit into your systems: the output needs to end up automatically in your existing software or workflow.
- You need to be able to explain it: you want to know why the model suggests something.
Our principles are practical: we use AI where it saves time or catches errors, every solution is explainable and testable, and your own team can maintain it. People stay in the driver's seat. Isatis is ISO 27001-certified and has delivered 100+ projects, which matters if your data is sensitive.
Where it often goes wrong
Most failed projects have the same causes:
- Starting with the tool instead of the question. A beautiful dashboard with no decision behind it stops being opened after a month.
- Starting too big. A project spanning all departments at once gets stuck on alignment.
- No owner after go-live. Models and integrations change along with your organization and need maintenance.
- Blind trust. A model can sound convincing and still be wrong. Keep a human in the loop and measure the output regularly.
The fix is rarely technical. It's about setting clear boundaries, starting small and continuing to improve after go-live.
Frequently asked questions
What happens to my data when I use AI?
That depends on the tool and the agreements in place. Find out where the data is processed, whether it's stored and whether it may be used to train models. Don't share personal data or sensitive company information with a service if you don't know the answers. With a custom solution, you help decide the setup and access yourself.
Which AI is good with Excel?
For quick questions about a spreadsheet, such as summarizing, writing formulas or describing trends, general-purpose AI assistants are usually enough. Always check the output. As soon as the same analysis comes back every week or needs multiple sources, a fixed connection to a data platform is more reliable than pasting things in by hand.
What is the difference between data science and AI?
Data science is the discipline of collecting, cleaning and analyzing data and drawing conclusions from it. AI is a collection of techniques that learn patterns on their own and use them to predict, sort or decide. Data scientists use AI as a tool, but data science also includes ordinary statistics and reporting.
Want to know where to start?
Not sure whether your data is ready, or which question is best suited for a first trial? Get in touch and tell us what you want to achieve. We're happy to think along with no strings attached.
About the author: Jeroen Penninx, Director and co-founder of Isatis.
