Impact
Efficient
On average ±2 hours of search per site replaced by direct access to information.
Reliable and scalable
Reduced risk of error and more complete insight through structured data. Applicable to hundreds of thousands of documents.
Speed
Example
Presented as a blueprint to all Dutch environmental services
About
The Omgevingsdienst Brabant Noord (ODBN) exists to do something seemingly simple: ensure that people can live, work and work safely. In practice, that means: Making decisions that have legal, social and economic consequences.
Every permit is such a decision. And more importantly, each new decision builds on everything that was previously established.
That makes the work complex. Not because the rules are unclear, but because reality piles up in documents, exceptions and history.
At the same time, the pressure is growing. Supervision must be increasingly risk-based. The demand for transparency is increasing, cooperation between authorities is intensifying, and information must increasingly be shared with the umbrella organization ODNL and other authorities.
That’s where tension arises: you have to make decisions based on everything that has ever been recorded, while that past is not always available at the time you need it.
“For us, it was crucial that the model not hallucinate. If something is not in it, it should just be reported. So far that has been captured well. Based on that, we said: this works. It brings out the good things.”
WHEN INFORMATION EXISTS, BUT YOU CANNOT ACT ON IT
The information was there. But not in a form that worked. What was an archive on paper turned out to be a reconstruction process in practice. Licensing officers and supervisors spent an average of about two hours per site searching through a PDF archive to get a complete picture. Not because the information was missing, but because it was scattered across documents that were never intended to be used coherently. Permits were not stored as data points. The archive was unstructured. And the full history was always necessary, because a new permit does not completely replace the old one.
As Elien Bisseling, Strategic Analyst of data-driven work at ODBN, describes it, “You spend several hours in that preparation to pull everything together to understand what’s going on there now. Every time.”
The consequences were visible. Inefficiency took time, but the risk of error and incompleteness was at least as great. Shadow administrations arose outside the system: unreliable statements on individual disks, which no one could be sure were complete and up-to-date.
At the same time, the external context was changing. The demand for data sharing from ODNL increased. Information not only had to be available internally, but also had to be reliably shared externally. Without a structural solution, ODBN would not be able to make that connection to national expectations.
The problem was not in the amount of information. The problem was that you couldn’t decide on it with sufficient certainty.
THE STEP FROM DOCUMENTS TO CHARACTERISTICS
The first reflex in these kinds of issues is to improve search: faster, smarter, better. But as long as permits remain documents, searching remains part of the job. Ruud Cools, principal data engineer at Team Rockstars, names what was really needed:
“A solution that is explainable, auditable and traceable, so that employees have reliable, consistent data.”
The breakthrough came when the perspective shifted. Not: how do we make documents more searchable? But: what’s in them that’s needed to make decisions? The answer turned out to be surprisingly simple: “Realizing that you can ask the same questions of all permits every time.” Each permit contains recurring elements such as date, type of application, location and environmentally harmful activities. In practice, these were scattered in legal text: implicit, inconsistent and difficult to compare.
The solution focused on making that information explicit. Permits were digitized and cleaned using advanced OCR, after which AI was used to recognize and extract data points as coherent characteristics that are comparable per permit and across permits. These were stored in a data mart, creating a new layer of information on top of the existing documents.
Reliability was central to this. The model quotes exclusively from the source and adds nothing. As Elien puts it, “We absolutely did not want him to come up with something himself. If something is not in there, he just says: it’s not in there.”
The approach was deliberately pragmatic. With a small team, a Proof of Concept was developed in ten weeks based on seventy permits, continuously validated by permit issuers and regulators.
What began as research, with the only expectation of a report, grew into a working prototype ready to move to production.
WHAT CHANGES WHEN INFORMATION IS READILY AVAILABLE
The impact is not in nuance, but in how the work fundamentally changes.
Whereas previously it took an average of two hours to get a complete picture per location, information will soon be available instantly, across hundreds of thousands of documents. That means not only speed, but above all security.
- Efficiency: ±2 hours of searching per site replaced by direct access to information
- Reliability: Fewer errors and more complete insight through structured data
- Speed: Working AI prototype in 10 weeks, ready to move to production
- Scalability: Applicable to hundreds of thousands of documents
- Role model: Presented as a blueprint for other environmental services
The impact is directly reflected in daily work. The capacity of 20 permit officers shifts to where it adds the most value: substantive assessment and decision-making.
Elien looks ahead: “It is now possible to have a dashboard for all locations that immediately shows what is licensed.” This also changes how ODBN performs its core task. Risk-oriented supervision becomes concretely feasible, because capacity can be focused on the locations where the impact is greatest.
Collaboration
The collaboration began not as a classic IT project, but as an experiment with one shared goal: This should work. “There were no hidden agendas. People wanted this project to succeed,” says Ruud Cools.
That openness made all the difference. By working together with ODBN’s substantive experts from the beginning, trust and focus were quickly established. Ideas were immediately tested against practice, so that the project did not get stuck in analysis, but resulted in a working prototype.
What began as research grew with it. Several follow-ups are now underway, from further development to production to new applications within other domains. The approach is also gaining traction outside ODBN, with six other environmental services seeking to build the same. As Elien points out, “Everyone did recognize themselves in the issues. And that the model does not hallucinate, that was one of the main points.”
ODBN is thus not only a user, but a pioneer of a new way of working. And that is precisely where the power of this cooperation lies: not only building something that works, but together showing how things can be done differently.
“What struck me is that even though they are not from our content world at all, they asked the right questions directly to our permittees and regulators. They really showed an intention to understand our business. Friendly, open, short lines of communication. And they were able to work completely independently.”
Want to know more?
Please fill out the form below and a colleague will contact you as soon as possible.
Contact
"*" indicates required fields
