The problem
This project aimed to replace a manual quarterly forecasting process required for bookkeeping with an AI-driven solution. The goal was to automatically generate forecasts using data from the client’s internal database while allowing manual adjustments to account for outages, supply shortages, and similar exceptional circumstances.
Industry: Finance
Technology: Machine Learning, Spark, Cloud
Our solution
The solution ran on a scheduled basis with a backend component utilizing Spark, Databricks, and time-series forecasting models. Users could interact with results through a dedicated web application to approve forecasts, route them to other team members for further analysis, add comments, and make manual adjustments as needed.
Given the requirement to forecast hundreds of different KPIs, we designed each model to best fit the specific behavior patterns observed during solution development. The models automatically adjusted for outlying periods, ensuring robust predictions across diverse metrics.
The impact
Upon deployment, the client reduced total time from forecast creation to acceptance from approximately three weeks to just a couple of days.
