Using HR Analytics to predict employee turnover in tech startups: an evidence-based HRM approach

Purpose This study seeks to test the effectiveness of Evidence-Based Human Resource Management (EBHRM) in tech startups. It compares two approaches to HR Analytics: the use of generic Human Resource Information Systems (HRIS) and bespoke statistical models built on firm-specific data to predict employee turnover. Design/methodology/approach A comparative case study of a small tech startup was conducted. Turnover predictions from the firm's HRIS were contrasted with custom-built logistic regression (LR) and decision tree models using data from 95 employees, assessing predictive accuracy and the actionability of insights. Findings The HRIS predictions were no better than random chance. The LR model was significantly more accurate. Both tailored models, using local evidence, identified specific, context-sensitive drivers of turnover that were entirely missed by the HRIS. Research limitations/implications Findings are based on a single case study with a limited dataset, and only two analytical techniques are tested. Practical implications Startup managers should critically evaluate generic HRIS analytics. The study suggests that cultivating an evidence-based mindset and training HR professionals to use simple statistical models on internal data is as important as investing in technology when employing HR Analytics to inform effective retention strategies. Originality/value This study advances research on EBHRM, HRIS and employee dynamics in startups by (1) clarifying that the “best available evidence” relies on contextual fit, transparency and interpretability; (2) proposing context-fit is a boundary condition for standardised HRIS analytics; and (3) pointing to the contextual idiosyncracies of tech startups as an explanation for why commercial HRIS may fail, as well as proposing transparency and simplicity as conditions for HR Analytics effectiveness in those organisations.