Model from Assistance to Work

Municipality of The Hague

Algorithm that can identify target groups within the group of people on welfare based on characteristics. The goal is: tailored services to work or participation.

Last change on 23rd of August 2024, at 14:26 (CET) | Publication Standard 1.0

• Publication category: High-Risk AI-system
• Impact assessment: DPIA
• Status: Out of use

General information

Theme

Social Security

Begin date

Field not filled in.

Contact information

[e-mail]

Responsible use

Goal and impact

The aim was to be able to determine an appropriate offer for people on welfare to work or participation pathways based on characteristics. The algorithm was a one-off analysis.

Considerations

The standard service is for consultants to make service choices based on interviews and data. Advice based on the algorithm could complement this.

By analysing welfare files, a one-off pilot was used to identify the hard and soft characteristics of benefit recipients that have contributed to job outflow in the past. Those characteristics were detected in the pool of benefit recipients and these citizens were invited for "the good talk" in an accelerated manner, so that they were dependent on benefits for as short a time as possible. This analysis only determined when a benefit recipient was invited, eventually everyone received an invitation to that interview.

Human intervention

The outcome of the algorithm will always be an advice to the consultant and can be overridden by the consultant at any time. Processes of which the algorithm is a part are performed manually by individuals and not by a system based on the algorithm. These processes can be stopped at any time by the management or manager of the Plus 500 programme.

Risk management

A DEDA track was conducted (The Ethical Data Assistant) to recognise potential ethical issues. A DPIA was also carried out to recognise privacy risks. During development of the algorithm, the project group monitored risks. The algorithm was not put into production. If it had gone into production, the algorithm's outcomes would be evaluated regularly to recognise any biases. Also, how the algorithm is applied by the consultant would then be evaluated.

Legal basis

The legal basis is implementation of the statutory task Participation Act.

Impact assessment

Data Protection Impact Assessment (DPIA)

Operations

Data

Assistance data, data on reintegration and participation pathways, on debt assistance and self-reliance. The output is not used as input in any other automated process.

Technical design

Data analysis using Excel and R. Decision tree. Cluster algorithm. Text mining. First, it is determined whether distinctive customer segments are possible with the available data. It then tests whether certain segments have a higher probability of outflow to work and whether there are other key focus areas for customised services. The algorithm functions well if it increases the probability of outflow to work or provides better services towards participation.