Discharge advice tool

Municipality of Amsterdam

Amsterdam residents with low income and savings can get municipal tax relief. The municipality checks their application. The algorithm analyses data and selects the requests that qualify for remission. The final decision always lies with the handler. The tool makes the process faster and easier for both employees and Amsterdam residents.

Last change on 2nd of October 2025, at 12:15 (CET) | Publication Standard 1.0

• Publication category: Impactful algorithms
• Impact assessment: DPIA
• Status: In use

General information

Theme

Social Security

Begin date

2025-07

Contact information

[e-mail]

Link to publication website

https://www.amsterdam.nl/belastingen

Responsible use

Goal and impact

The algorithm was created to make the tax remission process easier and faster for citizens and entrepreneurs. It predicts which requests are likely to qualify for remission anyway. This allows the municipality to work more efficiently and makes the service for Amsterdam citizens more modern and customer-friendly. Amsterdammers face the algorithm when they submit a request for remission. If the algorithm predicts that the request is likely to be approved, it gives this advice to the handling official. The latter takes this advice into consideration. If the advice is followed, the citizen does not have to send additional information. This makes the process faster and simpler for the citizen. If the algorithm cannot give any advice, the request is assessed by an employee in the normal way. The algorithm ensures that citizens have less administrative burden and receive an answer faster. For the municipality, this means a more efficient process and less work. An employee always makes the final decision, so the process remains reliable. The algorithm is used within the municipality.

Considerations

The algorithm makes the waiver process faster and simpler. Citizens need to provide additional information less often, saving them time and effort. For the municipality too, this means less work and lower costs. The process becomes more customer-friendly and efficient, while an employee always makes the final decision. This ensures reliability. One drawback is that the algorithm can sometimes make a mistake. If an employee copies this, it can lead to an unjustified grant of remission. This means a small financial loss for the municipality, but it is offset by the time savings and better service. Through monitoring, retraining and transparency, we mitigate the risks . The advantages clearly outweigh the disadvantages, justifying the use of the algorithm.

Human intervention

The algorithm provides an opinion that helps staff assess waiver requests. The algorithm gives a positive advice (assign) only if it deduces from the request that it is likely to be approved. In doing so, it indicates how likely this outcome is. The clerk can make the appropriate decision based on this advice. All other requests are assessed by a staff member as normal. Employees use the advice to work faster, and always make the final decision. The results of the algorithm are checked by regular monitoring and sampling. In addition, the algorithm is regularly updated and re-trained with new data to ensure it remains reliable. Sometimes the tool does not pass on a positive opinion. The final outcome of these cases is used both to further train the tool and to ensure employees stay sharp.

Risk management

Risks are managed in various ways. Errors only lead to an unjustified award, which is a small financial loss offset by time savings and better service. The algorithm is regularly retrained with new data to remain reliable. Its performance is continuously monitored and randomly checked to detect errors quickly. The algorithm is included in the algorithm register for transparency. A certainty level is applied so that the algorithm gives advice only if it is very sure of the advice. If the algorithm does not give advice, a staff member fully assesses the application. This keeps risks down and the process reliable.

Legal basis

The legal basis for the process of remission of municipal taxes is laid down in the Invorderingswet 1990 and its Implementing Decree Invorderingswet 1990. These laws and regulations specify the conditions under which citizens can qualify for remission of municipal taxes, such as waste collection or sewerage charges. Municipalities are obliged to assess requests for remission based on the applicant's financial situation. The algorithm is used as a tool in this process: it advises on the likelihood of granting. The final decision is based on the legal frameworks and is made by an employee.

Links to legal bases

Recovery Act 1990: https://wetten.overheid.nl/jci1.3:c:BWBR0004770&hoofdstuk=IV&afdeling=3&artikel=26&z=2025-01-01&g=2025-01-01

Link to Processing Index

https://www.amsterdam.nl/privacy/verwerkingsregister/verwerkingsregister-avg/

Elaboration on impact assessments

The final decision is always taken by an employee, ensuring human control. Even if wrongful waiver is granted, the citizen's right is not affected. Should there be an impact at all, such as an unjustified grant, this is precisely a recognition of fundamental rights by design. The algorithm has no direct impact on fundamental rights such as privacy or non-discrimination. It will be included in the algorithm register for transparency. Through monitoring, retraining and an escalation procedure, risks are managed. Control measures have been taken in the areas of input (Privacy by Design, Bias by Design), throughput (regular retraining and monitoring) and output (only positive opinions, sometimes not disclosed to employees). As a result, the impact on human rights is very limited and human control is ensured. A BIAS analysis has been carried out. This analysis focuses on detecting and reducing possible biases (bias) in the algorithm. The aim of this analysis is to ensure that the algorithm works fairly and objectively and does not disadvantage groups. The results of the BIAS analysis were taken into account in the further development of the algorithm. Together with the DPIA, this ensures that the algorithm is used responsibly and transparently. An ethical leaflet is planned.

Impact assessment

Data Protection Impact Assessment (DPIA)

Operations

Data

The algorithm uses the following data: age of the applicant, financial data such as rent standard, income, wages, housing costs and assets (also over several years, e.g. since 2017 or the last 5 years), and historical data, such as how often remission has been applied for, granted or not granted in recent years (e.g. the last 2-6 years). This data comes from legitimately authorised sources and own (historical) tax information. This is the same information that is already used when assessing waiver requests, but combined and in larger quantities. All data is processed according to Privacy by Design and within applicable laws and regulations.

Links to data sources

https://www.bidn.nl/belastingen: De gegevens komen van het Bureau InformatieDiensten Nederland en de historische gegevens zijn gebaseerd op eerdere beslissingen van de afdeling Belastingen.

Technical design

The model predicts whether an application for remission is likely to be granted. To do so, it uses historical data from previous applications. Based on this, it learns to recognise patterns associated with granting. For a new application, the model calculates a probability percentage indicating the likelihood of granting. This percentage, along with the main determining factors, is displayed as an advice to the practitioner. The model uses an AI prediction method based on machine learning.

External provider

VyZyr (https://www.vyzyr.nl/)