Job placements – support programmes don’t benefit everyone equally.

The digital transformation raises the question of whether data-driven decision-making models can improve job placements. Michael Lechner from the University of St. Gallen and his team looked into this question as part of National Research Programme 77.

The goal was to use data and algorithms to better connect unemployed people with support services. To achieve this, the team used the latest causal machine learning techniques – i.e. machine learning techniques that recognise cause-and-effect relationships – and developed an algorithmic recommendation system. This system suggests tailored support programmes for unemployed individuals, such as further training, job application coaching or subsidised employment, all designed to improve their job prospects.

The most important findings

Machine learning enabled the research team to determine the extent to which each unemployed person benefited from the various programmes, thereby measuring the effectiveness of one aspect of Swiss labour market policy.

The analysis revealed clear differences. Some support programmes significantly improve certain groups’ chances of finding employment or earning higher incomes, while having little impact on others. For example, one group benefited greatly from a particular training course, while another responded better to a different programme. These findings highlight the importance of differentiated approaches, namely that what works well on average is not necessarily equally effective for everyone.

Based on these detailed results, the researchers developed an algorithmic recommendation system that suggests the programme most likely to benefit newly registered unemployed people with certain characteristics. The system also considers practical constraints such as limited course places and budget constraints.

The researchers also considered it important to make the recommendations transparent. The process of ensuring the best possible recommendation for a given unemployed individual is presented in the form of a policy tree so that users can clearly see the criteria on which the recommendation is based. This ensures interpretability and enables decision-makers to apply the recommendations with confidence.

Relevance for policy and practice

These findings provide promising starting points for labour market policy and employment services. With data-driven decision-making algorithms, employment agencies could implement measures more precisely and efficiently. Such tools would support the selection of suitable programmes for individual job seekers, ensuring that limited resources such as course places or subsidies are allocated in ways to achieve the greatest possible benefit.

Nevertheless, the findings of this NRP 77 research project clearly show that some prerequisites must be met for its successful implementation. Alongside comprehensive data and a suitable IT infrastructure, it requires competent specialists who understand how the algorithms work and how to interpret their recommendations correctly.

Three main messages

1. Use causal machine learning to better understand which public programme works best for whom.
2. Use data-based decision algorithms to improve the allocation of individuals to specific programmes.
3. Ideally, there should be a continuously updated, integrated data-decision pipeline, i.e. a continuous flow of data and a semi-automatic update of the evaluation results from which policy recommendations are derived.

You can learn more about the researchers’ methodology and background to the project on the NRP 77 project website:

Further NRP 77 research projects on the topic of “Digital Transformation” can be found here: