Algorithms: Prevention and Compliance

A person who shares a household with 1 other adult gets less in AOW pension than if they live alone. We can check whether a person who is registered in our system as ‘living alone’ is more likely living in the same house as another adult. We do this by using the SWAN model. This model can predict the probability that a person ‘living alone’ is in fact living with another adult. The SWAN model is an algorithm.

Our Prevention and Compliance Department can pay you a visit at home in order to check whether you are living with another adult. These are individual visits. We use the SWAN model here to select which visits we need to make in order to keep the number of face-to-face checks to a minimum.

The SWAN model helps us minimise risk by more easily identifying situations in which people may be receiving benefits that they are not entitled to. We can then take action to make sure that the amount you have to repay stays as low as possible. Our compliance officers can also work more efficiently. By using the model, we can save money on benefits and on the amount of work we have to do.

The model learns continuously from earlier instances where it turns out that a client received more than they were entitled to, and had to repay money to us as a result. Data from the past allows the model assess the risk that a particular client may be receiving more than they are entitled to.

This algorithm is not prescriptive. This means that it cannot take an automated decision. Instead, the algorithm produces a selection of probable cases which are then assessed by our compliance officers. The compliance officer looks at all the available data relating to the client's living situation.

We always check whether our models comply with the necessary rules on ethics and controllability. We use the rules set by the Netherlands Court of Audit and our own internal rules.

The development of the algorithm was monitored by expert parties both within and outside our organisation. They ensured that the algorithm was developed securely and with great attention to detail.

We use the client data we have on our files, for example: 

  • the client's current and/or former address 
  • their domestic situation 
  • age 
  • sex