Leveraging predictive analytics to modernize labour market support for the Lithuanian Employment Service

Užimtumo tarnyba is the primary governmental institution responsible for labour market policy implementation, employment facilitation, and support for job seekers nationwide. It manages a large database of registered individuals and aims to deliver more personalized, high-impact interventions to reduce unemployment.

To modernize labour market interventions, the client sought to move beyond manual assessments and develop a data-driven 360-degree statistical profiling system. The objective was to create a model that could accurately predict the probability of a job seeker finding employment within 12 months of registration. Such a tool was essential to help the agency prioritize resources and provide specialized support to those facing the greatest barriers to employment.

• Phase I: Best Practice and Scientific Review

Our work began with an extensive audit of international best practices and scientific literature on statistical profiling. We analysed successful models from other European countries to identify the most effective methodologies and variables for predicting long-term employment success.

• Phase II: Analysis of Data Sources

As part of the project Civitta team performed a comprehensive analysis of available data sources to identify their relevance and suitability for predictive modeling. This process included mapping all internal and external data inputs, understanding their structure, and evaluating their consistency, completeness, and reliability. Based on these insights, the most robust and informative datasets were prioritized, ensuring that only high-quality and fit-for-purpose data would be used in the subsequent stages of developing the prediction model.

• Phase III: Exploratory Data Analysis

Before building complex models, we conducted exploratory data analysis. This allowed us to observe primary trends in the labour market data, identifying the key drivers such as age, education, and region that most significantly influence a person’s likelihood of finding a job.

• Phase IV: Model Concept and Development

This was the core stage in the project. Our goal was to develop a machine learning model to estimate probability for a given person to find a job. We developed and tested multiple advanced statistical models. This iterative process involved selecting the most accurate model based on precision metrics and ensuring it could handle the complexity of the national labour market while remaining interpretable for everyday use by agency staff.

• Phase V: Model Implementation

The final phase involved integrating the ready-to-use model into the Employment Service’s internal systems. This transition from a concept to a functional tool ensures that employment specialists can use data-driven insights in their daily operations to support job seekers more effectively.

Civitta delivered a centralized, modern predictive solution that significantly improves the agency’s ability to target its interventions. By replacing generic assessment methods with a statistical profiling model, the Employment Service can now identify high-risk individuals early, optimize budget allocation for training programs, and accelerate the transition of job seekers back into the workforce.

The system enables a shift from administrative reporting to strategic, result-oriented decision-making. Enhanced data transparency and predictive capabilities provide the organization with greater control over its labour market outcomes, ultimately strengthening the national economy.

• By predicting the probability of employment success within 12 months, the agency can identify high-risk job seekers immediately upon registration and deploy targeted support before long-term unemployment occurs.
• Statistical profiling allows for more precise budget management, moving away from “one-size-fits-all” programs toward strategic interventions that maximize the ROI of labour market policies.
• The success of the delivery was anchored in a rigorous 5-stage data science methodology, ensuring the model was grounded in both scientific research and operational reality.