PredictiBank Qorus Reinvention Awards - Europe 2025 - Winner

Our goal is to develop an algorithm that utilizes artificial intelligence and advanced analytical methods to automatically predict quarterly targets given to branches and their employees, and distribute the predicted targets according to the top-level budgets set by the Bank. This algorithm transforms the manual process into an automated one for all metrics found in the performance scorecards of branches and their employees. As a result, branches and their employees are encouraged to set explainable, achievable, accurate, challenging, measurable, and appropriate targets by using dynamically trained models. This encourages the Bank to increase its competitive advantage, market share, and penetration rate in the market.

Our advanced analytical approach for branch target distribution is an automated tool consisting of three stages. The first stage is model training, where 69 different models are regularly trained based on branch and employee data. We created a variable pool specific to each model, using variables such as historical performance of branches and their employees, seasonal variables, and variables reflecting penetration in relevant items. Model training is done using the selected variables from this pool (varies based on the model). The average MAPE performance of these models is 29.7%. The models are regularly trained every six months. The second stage involves predicting target at the branch and its employee level using the models trained (fed by stage 1) during the budget distribution periods (usually quarterly). The third stage involves a distribution algorithm that distributes the Bank's top-level budgets to the branch and its employee level. These three stages work sequentially and generate outputs to the budget control team. In addition, the three stages are designed to adapt to potential changes in distribution periods in the future. In other words, when transitioning to monthly or semi-annual targeting in the future, the models can automatically generate the outputs. This flexibility allows the system to adapt to changing environment and estimate target predictions in different periods. As a result, the Bank's target distribution process becomes more flexible and scalable.

With this framework, we automated the branch target distribution process and reduced the time required by 50% compared to the previous method. By making the targets transparent and understandable, we were able to reduce the number of complaints from branches and their employees by 80%. Additionally, thanks to the accurate and challenging targets set, we observed additional sales results in indirect financial gain.