Human Resources Predictive Analysis

HRPA is our solution to the turnover challenge that many companies face. By leveraging company data, we build and deploy state-of-the-art machine learning pipelines to detect potential leavers months before their resignations. This information is enriched with a layer of explainability and delivered through effective dashboards for easy consultation, allowing our clients to understand the drivers of this phenomenon.

Scenario

A multinational company experiencing a significantly higher turnover rate than its reference field: ~10% vs. 5%. Despite large investments, their retention policies have only been mildly successful, leading management to question their approach. We proposed testing whether our AI model could outperform their HR department’s capabilities. This is how HRPA was born.

The main goals are threefold:

1. Provide a reliable estimate of who is more likely to leave the company to funnel resources where they can make a difference.
2. Understand the structural factors driving the phenomenon to correct them.
3. Test different retention strategies for each person to apply tailored retention policies.

Method

As an agile company, we agreed to deliver a proof-of-concept (PoC) to our client to give a first estimation of potential leavers with minimal effort. To achieve this, we analyzed the available data sources, considering factors such as time depth, cleanliness, significance, etc. We focused on the smallest perimeter that could give the best results with reasonable confidence.

The data was processed with a customized pipeline, making the most of it. Using our AutoML framework, STEAM Engine, we evaluated thousands of models, checked if they could be combined into an ensemble with superior performance, and tuned the ensemble to maximize its effectiveness.

We also applied explainable AI techniques based on cooperative game theory to provide an initial estimation of the most relevant factors for the model’s predictions. Finally, the data was presented in a dashboard created alongside the client, allowing their HR department to benefit from our work.

The following phases of the project focused on:

• Deploying the pipeline in production;
• Adding more data sources;
• Fine-tuning the model;
• Greatly enhancing feature explainability;
• And more…

Comprehensive HR Metrics for Enhanced Workforce Analysis

The following categories help provide a comprehensive and detailed view of the dynamics that can influence employee turnover and overall company performance, thereby facilitating the creation of predictive models and the adoption of more effective management strategies.