The Responsible Organisation

The Norwegian Labour and Welfare Administration’s (NAV) is a public agency with the mission to provide social and financial security and to facilitate the transition into work and activity for individuals who are currently unemployed or not engaged in any professional activities. To fulfil its objectives, NAV oversees a range of social welfare programs including supported employment, joblessness compensation, work evaluation allowances, illness benefits, retirement pensions, monetary social support, parental leave pay, child allowances, and care-related cash benefits. NAV provides services, grants, and benefits to both individuals and employers. From a holistic approach to social protection and unemployment, NAV cooperates closely with employers, businesses, municipal, health and education sectors, other state agencies, and voluntary organisations.

The problem

In NAV, the Central Government and municipalities co-operate for the provision of a single gateway into public labour and welfare services. A key task for NAV is to help citizens get back or into work as quickly as possible. To do so, NAV needs to know who needs follow-up, and how to arrange it. Many people register with NAV every year, and there is the need to quickly identify their needs. Providing assistance to all the population can be resource demanding and can lead to a long waiting period. 

As timing is fundamental in these situations, the need to find quick solutions pushed the administration to look out for innovative tools that would support this process in a cost-efficient way, enhancing the quality of the service as well, specifically to automate this process and make more effective decisions.

The solution and its implementation

NAV implemented a rule-based algorithm that allows to classify jobseekers’ profiles into different categories every time that a person registered as a job seeker requests assistance to find a job or requests unemployment benefits. Individuals can register as jobseekers, which is a requirement for applying for unemployment benefits, either independently or with the assistance of a NAV employee.

NAV categorises users according to the level of  help it is assumed that the person needs (NAV Act, article 14a, 2009). The category in which a case is classified affects the kind of follow-up or effort that NAV has to devote for them. There are different types of effort, which are described in the table below (NAV, 2024):

From 2023, the NAV National Follow-up Unit (Nasjonal oppfølgingsenhet - NOE) introduced the rule-based algorithm that support this decision-making process. Established in 2023, the NOE is a virtual unit that follows up standard users between 30 and 59 years old throughout the country. All processes and interactions with the users are carried out by messages, or if needed, video-phone calls. It has approximately 30 caseworkers who are located across Norway. Each caseworker is supervising several users/applicants and keeping track of the evolution.

The system creates a recommendation to the applicants’ supervisor taking into consideration applicants’ age, education, stated health challenges and recent work history. The NOE receives cases from the algorithm and from the local offices, and the caseworkers assess if the assumed standard users are indeed standard users or should be transferred to the local office because of higher assistance needs – i.e. situational efforts programmes.

• Generally, “Standard Effort” is proposed to people between 18 and 59 years of age with at least completed primary school, without health challenges, who have worked for at least six of the last 12 months, and who do not state any other challenges in applying for or staying in work.
• Conversely, “Situational Effort” is proposed for those under 18 or over 60, who have not received approved education, or have not been in work for at least six of the last 12 months.
• On the contrary, for individuals reporting significant health problems or other barriers to employment, a work capacity assessment is conducted by NAV caseworkers. Interestingly, a notable proportion of these cases are subsequently categorised under "Standard Effort" rather than more intensive support categories. This often occurs because many jobseekers may mistakenly report their abilities during registration or because caseworkers, upon review, deem the identified challenges as not significantly impacting their ability to work. The cases identified as requiring more comprehensive support are classified for “Especially Adapted” or “Permanently Adjusted” efforts.

When first introduced, the algorithm achieved an accuracy rate of around 33% in correctly classifying cases for the "situational effort" compared to NAV caseworkers’ decisions. A year later, this increased up to 50% of the cases’ decisions, signalling an increase in the accuracy of the algorithm from the caseworkers’ perspective.

An important reason for partially automating processes is to reduce the time required on the simple matters and dedicate NAV resources to more value-added tasks such as conducting in-person meetings with NAV’s beneficiaries to provide tailored assistance to cases falling within the “situational effort type” or higher. Ensuring the right level of support from NAV has in fact proven to be important for the unemployed to find work quickly. According to NAV’s researcher Espen Steinung Dahl, results show that a larger number of unemployed people receive decisions about situational efforts, and therefore tailored support from NAV after the algorithm has been put into use.

Expected benefits

The introduction of the profiling algorithm has shown to bring multiple benefits to the administration.

• Faster processing procedures: With the implementation of the algorithm, the NOE has an active user base of approximately 18,000 individuals. On average, the NOE receives about 1,000 new cases each week, meaning 52,000 per year since the introduction of this system in 2023. 
• Online platform enables faster case management: the solution enables a more efficient approach to case management, as it supports an effective identification of cases that can be managed through online support and meetings by caseworkers. Nonetheless, NAV still prioritises direct assistance for certain user groups. Specifically, standard effort cases involving individuals under 30 and over 60 are routed directly to local offices.
• Resources and assistance time are increasingly focused on who need it most: the applicants that need tailored assistance are helped faster and further resources are focused on these cases, including in-person support activities, since only “situational effort” cases or higher are now handled by local offices.  

Main challenges

The challenges related to the implementation of the algorithm are however multiple.

• Data protection regulations: as of now, there is still some discrepancy between the algorithm results and the caseworkers’ evaluations, due to their subjective evaluation based on professional experience. To improve the algorithm accuracy with a more advanced machine learning tool and reduce the need for human revision and adjustments, additional personal data from applicants should be required. However, data protection regulations pose a challenge.
• Other tasks are still carried out manually: all communications and follow-ups to applicants are made manually and can be time demanding, leaving room for improvement for further automation of repetitive tasks.
• Unclear scalability across NAV personnel: since all the caseworkers have long experience in the local offices and know how to identify issues in each of the jobseekers, it is unclear how the solution may retain its accuracy if scaled to teams of new case workers without previous experience, for which they would not be able to evaluate or spot judgment errors from the algorithm.

• NAV official website
• Algorithms gave unemployed people decisions about more help from NAV
• Does user profiling lead to more accurate work-oriented follow-up?