Company: KocDigital A.S., Istanbul, Turkey
Nomination Category: Best Use of People Analytics
Loss of skilled personnel is a leading cost item for any company and one of the most important issues for Human Resources functions.
Estimating the risk of personnel resignation in advance, and taking preventive measures by identifying potential reasons for resignation is now possible, thanks to the turnover estimation project implemented through the cooperation of KocDigital and KocSistem HR.
KocSistem HR believes in the extremely crucial role of analytics in the development of a data-driven decision-making culture with respect to both strategic HR decisions and the operational processes on the journey of digitalization. In its quest for “Integrated Digital Processes & Sustainable Analytics”, KS HR brought about a truly innovative dimension in its endeavors, by achieving the first case of highly accurate turnover estimation in Turkey.
The first step of the analytics road map at KocSistem has been to delineate the analytics strategy for HR and the “Create Engaged Employee & Effective Company with Analytics in HR” vision, with the involvement of senior management. With a view to deploying this vision analytics trainings were organized for the whole HR team.
The workshops held after the trainings led to the identification of 10 use cases to generate added value through the analytics perspective in HR. ‘Turnover Estimation’ was selected as the highest priority among these 10 cases.
The loss of trained work force in the tech sector, where competition is intense and knowledge is the most important element of capital, can inflict so much damage to affect even the companies’ competitiveness in the market. Moreover, resignations would bring about extra hiring, onboarding and training costs out of HR budgets, in the context of replacing the lost personnel.
That is why estimating the risk of resignation of personnel with high performance and strong potential, followed by action required for preventing turnover before it occurred, helps firms keep costs manageable and protect their intellectual capital.
The project was implemented by using Agile Methodology with a team comprising KocSistem’s HR Process owners and KocDigital Data Science specialists,
All data kept in corporate systems were analyzed and collected with a framework of an employee data model which consists of 64 distinct types (e.g. personnel demographics, leaves, trainings, qualifications, personality inventory, work-life balance, talent management, feedbacks, participation in social events) for 1800 employees, for the past 3 years.
The Data Science team then developed a model estimating resignations, using randomly selected 70% of the data thus gathered. The remaining 30% of the data, used as test data, allowed determining the accuracy of this model. Additionally, the analyses performed led to the identification of the factors with an impact on turnover.
The advanced analytics techniques (SMOTE, Stratified Sampling, Hyperparameter Tuning, Repeated K-fold X-Validation, XGBoost, Random Forest) employed with respect to turnover –a field based on human behavior and inherently difficult to estimate, not to mention the limited volume of data applicable– allowed highly accurate results. In addition, to enable reproducibility of the outcomes, all of the data used in the project were integrated in a central data warehouse which decreased the cost of data collection.
At KocSistem, the staff members who are deemed potential candidates for resignation, starting with the highest risk group, are kept under close monitoring through personal meetings with the relevant HR-BP and the employee’s supervisor, within 1 month.
Then comes remedial action with a concrete effect on the staff member, with respect to the factor which poses a risk in her individual case. In case the personal meetings confirm the risk of resignation, churn management is introduced, supported with meetings with the Chief Executive, coupled with support for the career track of the employee, through rotation referrals on KocKariyerim, with a view to keeping that employee in Koc Group.
-Model developed is able to estimate resignation/continued tenure probabilities for employees with an accuracy rate of 93%
-Focusing only staff members who will resign model is able to accurately estimate individuals to resign, 70% of time
-18 major factors with highest level of effect on Turnover (promotion, training attendance rate, age, wage details, leaves)
-Model assesses resignation risks for specific individuals and factors which lead to resignation. In terms of risk levels assessed, KS personnel are categorized in 4 groups: high-risk, medium-risk, low-risk and safe zone.
-Model is run each month based on current data to allow HR and supervisor responsible with relevant organization monitor risk scores and list of personnel in current risk categories on HR Dashboard.
-Action plans were implemented for 120 employees who were in applicable risk group.
-These efforts are lead to a 2% reduction in unexpected turnover, which will translate to savings of TRY 300K.