The Netherlands is investing heavily in the energy transition. Through schemes such as the ISDE (Investment Subsidy for Sustainable Energy), SEEH (Subsidy for Energy Saving in Owner-Occupied Housing) and municipal funds, households and businesses can recover thousands of euros on investments in solar panels, heat pumps, insulation and charging infrastructure. The application process is largely digital: applicants upload photos of the installation as proof that the measure has actually been carried out.
That sounds sufficient. But it also opens a door to fraud that is difficult to detect manually.
The subsidy fraud problem in practice
The most common forms of subsidy fraud in energy measures are not spectacular. They are deliberately subtle.
Photos from a different address. An applicant claims a subsidy for solar panels on their own roof, but the uploaded photos were taken at a neighbour’s home, a family member’s property, or even a random house found on Google Street View. Without a site visit, this is difficult to verify.
Reuse of previously approved photos. The same set of installation photos is submitted with multiple applications: sometimes by the same applicant for the same property (double-dipping), sometimes through a network of fraudsters exchanging each other’s photos.
Installer-facilitated fraud. In rarer cases, an installer carries out work only partially, or not at all, and provides photos from elsewhere as proof for the customer. The subsidy money is shared.
The scale of the problem is hard to quantify precisely, because detecting it is labour-intensive. RVO, the implementing agency that manages the ISDE, processes tens of thousands of applications per year. Each file contains multiple photos. Manually inspecting every file is simply not feasible.
Why EXIF metadata is insufficient
The obvious first step is checking the GPS coordinates embedded in the EXIF metadata of a photo. Many smartphones record these automatically when taking a picture.
But EXIF verification has fundamental limitations in subsidy checks.
First, EXIF data is trivially editable. Dozens of free tools are available that modify embedded location data. A fraudster who knows the responsible agency checks coordinates simply changes them accordingly.
Second, EXIF data is often absent. Photos taken via WhatsApp, downloaded from a website, or shot with an older camera typically contain no GPS coordinates. Using the absence of an EXIF location field as grounds for rejection would block too many legitimate applications.
Third, EXIF data is not proof of reality, only of what the device recorded at the moment of capture. Inside a building, GPS inaccuracy can run to hundreds of metres.
Visual photo geolocation, which determines a location based on what is visible in the photo, provides an independent verification layer that does not share these limitations.
How AI geolocation strengthens subsidy verification
GeoPin analyses the visual content of a photo and compares it against its detailed index of Dutch addresses and environments. For a photo of a roof with solar panels, this means: does the system recognise the roof structure, the facade composition, the window layout, the surrounding buildings, the street layout?
The result is a location prediction with an accompanying confidence score and the most likely GPS coordinates.
In subsidy verification, this produces three direct applications.
Address verification. The predicted location of the photo is compared with the address in the application. Does the photo visually match the facade, roof and surroundings at that address? A large discrepancy is a signal for further investigation.
Deduplication detection. The same photo submitted with multiple applications is recognised both by visual similarity and by identical geolocation results. A set of photos that already appeared in an approved file cannot serve as evidence again.
Batch screening during application peaks. After campaigns or expiry deadlines for subsidy schemes, large numbers of applications arrive in a short period. Automated geolocation screening processes hundreds of files per hour, so that human inspectors only need to review the flagged cases.
Integration with the existing application workflow
A practical implementation does not need to replace the existing application environment. The GeoPin API integrates as a verification layer between the upload moment and the final assessment.
The workflow looks like this:
Step 1: Upload. The applicant uploads photos via the existing portal. Nothing changes on the front end.
Step 2: Automated scan. The submitted photos are passed to GeoPin via the API. Per photo, the system returns a location prediction, confidence score and an indication of whether the photo has been seen before in the system, all within a few seconds.
Step 3: Routing. Applications with high agreement proceed normally. Applications with a significant location discrepancy or low confidence are flagged for manual review.
Step 4: Human assessment. The reviewer sees the application together with the geolocation result, the difference between predicted and stated address, and any matches with previous applications. That is concrete input, not a vague suspicion.
Exterior photos versus installation photos
A valid question: geolocation works well for exterior photos, but installation photos of solar panels or a heat pump in a utility room sometimes contain little outdoor context.
That is correct. Subsidy packages typically contain a mix of photos: overview shots of the roof or facade (high geolocation value), detail photos of the installation (low geolocation value) and photos of energy labels or measuring equipment (no geolocation value).
The overview photos are most valuable for verification, and they are present in virtually every subsidy file. Installers and applicants are accustomed to taking them as evidence of the situation before and after the installation. The system only needs to find one or two photos with sufficient outdoor context to perform a reliable location check.
Why this is urgent now
Subsidy flows for the energy transition are growing, not shrinking. More money means greater appeal for fraud. At the same time, the implementing agencies that process applications have not grown proportionally in capacity.
The pressure to verify more efficiently is therefore structural. Automated photo geolocation is one of the few techniques that offers scalability without requiring a human inspector for every file. The system operates at location level rather than applicant level, making it more privacy-friendly than biometric or profiling techniques.
For implementing agencies looking to strengthen their fraud-resistant processes without extending processing timelines, this is a concrete technology option that is deployable today.
GeoPin provides photo geolocation specifically optimised for the Netherlands, including API access for automation. Read more about how GeoPin works or explore API integration options.