Outsource your data analysis

You give an external provider a question you want to answer using your own data. They carry out the analysis and explain the findings. It might be a one-off study, such as finding out why customers leave, or an analysis your team can repeat afterwards.

It is one of the AI applications we build. If you want to compare providers first, see our overview of data analytics consultancies.

When you outsource, you pay for an answer to a specific question. When you hire, you pay for capacity: someone who works alongside your team and follows its priorities. A seconded specialist from Data Science Lab is employed by us.

If you hire a freelance analyst yourself, pay attention to the Wet DBA. Since 1 January 2025, the Dutch Tax and Customs Administration has resumed full enforcement of the rules on false self-employment. The assessment depends on how the work is carried out in practice.

Only what is needed to answer the question: an export from your ERP or CRM, an Excel list, or read access to your data platform. Access to production systems is rarely necessary. Just as important is someone who knows the source and can explain what each field means.

If no one knows how a column was populated, we record that as an assumption in the explanation. That way, you can see what the findings are based on later.

When we analyse personal data for you, we act as a processor. The European Data Protection Board says that a processor acts solely on the instructions of the controller and that this relationship must be set out in a contract. You remain responsible, including for the processor’s compliance.

We ask only for the fields needed to answer the question. According to the Autoriteit Persoonsgegevens, pseudonymised data usually remains personal data. Your privacy officer or legal adviser assesses whether you are permitted to carry out the analysis.

We start with exploratory data analysis: examining the data before drawing conclusions, identifying outliers and anomalies, and testing assumptions. We then choose the techniques the question calls for, from comparing groups to regression or a model that detects anomalies, as we did for GVB with logged hours.

We work in Python and SQL, using notebooks on your own platform where possible, such as Databricks or Microsoft Fabric.

Often, you cannot know, and we say so. Two figures that move together may both be influenced by a third factor, such as season, region or customer group. One common pitfall is Simpson’s paradox: a relationship seen across a whole group can disappear or reverse when you break the group down.

That is why we break down the findings and explain what we could not rule out. If you want to demonstrate a cause, an experiment with a comparison group can help.

A plain-language explanation of the findings, with the supporting evidence, the definitions we used and what we could not establish. The code is stored in your own repository so your team can repeat the analysis with new data.

If the analysis needs to run regularly, we can turn it into a repeatable pipeline through data engineering, or create a report through business intelligence.

When the question is not yet clear, or no one is ready to act on the findings. In that case, start with a maturity assessment. If your data is so scattered that every analysis starts with manual work, address that first in your data foundation.

If the same question keeps coming up, regular reporting through business intelligence is more useful. If your question is about who your customers are and the value they generate, see customer analysis.