Client story: Regulatory authority

Streamlining the case handling process in social services

Creating more comprehensive, safe, and accurate assessments of social service cases

This municipality in Stockholm County covers a third of the region, and has over 60,000 citizens. One of its five offices covers social services.

The social services are organized into four departments, one of which administers and investigates concerns raised by citizens. This department has about 70 employees, seven of whom perform each case’s initial assessment. The department wanted to explore the use of AI models, while, importantly, remaining ethical and compliant.

Download the case

Download the PDF-version of “Streamlining the case handling process in social services” client story

The challenge

The applications which the social services department had in mind are crucial both socially and in terms of data protection. Using modeling techniques or AI in such cases increases the need for governance. Currently, many processes are handled manually, including the sorting of cases raised by citizens. This is an area that can benefit from using machine learning to signal to civil servants where to draw their attention and which case to process faster to avoid a huge backlog including critical cases. That the cases take the individual civil servant a good deal of time to assess properly plays into the business case.

In many cases, the time it takes to process information in a municipality is critical, with any potential delays preventing action being taken in time to help. However, speeding up this process is not possible without minimizing risk through compliance governance, and controlling algorithms and models for data processing.

The municipality wanted to ensure that the models they produce are to the highest standard ethically and follow all laws that relate to data privacy, compliance or model/AI governance. They wanted not only digital controls implemented but also to be able to add personal comments and notes.

The solution

A Natural Language Processing (NLP) model was trained on historical data to support civil servants by assessing and prioritizing cases based on their content.

The model is embedded in Grace. That means that all the restraining procedures of the model are logged. Changes to the code are documented and the model performance is reported. In addition to this, the municipality can add comments and document decisions in assessments which in themselves generate metadata. This metadata can then be used to steer and control any development work that might be performed by external development teams or consultants.

When applying AI, it is important to ensure that the AI model does not have access to personal data. To ensure this, the system checks for personal identifiable information (PII), which is then excluded and not used.

In areas where we see certain types of cases and reports being extraordinarily similar, we can manage our resources and meet those who need our help the most, significantly earlier. Thus, spending time on the most value-creating areas of the social service department.

CIO of the municipality

The results

The solution gives the municipality the perfect starting point to either build machine learning models that can be implemented in their organization, or to use the governance engine to monitor other IT processes already in place such as robotic process automation.

With this AI model in production, we will reduce the administration for civil servants who will instead be able to work with what they are trained to, namely meeting people.

Department Head

Project highlights

  • The Grace platform ensures a complete audit trail and history, combined with Impact Assessments
  • An increase in the quality of information and decision-making data
  • A significant reduction in administrative tasks

About the project

The project is run in collaboration with Vinnova, Valcon Consulting, AKOA, and the municipality. 2021.AI covered the AI area, including use case and model development, the continued platform support, and retraining of the model.

Ready for streamlining case handling processes?