Presently, every company has access to huge sets of information and high-capacity processing power. Simultaneously, tools are widely spread and capable of answer multiple purposes in complex environments. Specifically for forecasting, there are multiple software prepared with several algorithms, pre-processing methods and advanced UI – nowadays, the challenge of having a forecasting module relies more on reliance, adaptability, and deployment.
As a result, the difficulty falls on setup and adaptation to business requirements. More important than connecting the plug is knowing how to exploit and adapt all the modules provided. Faced with skeptical teams and poor results, successfully deploying a well-establish forecasting tool has been a tough challenge.
To improve the results and foster team’s confidence, the approach followed 3 key structures:
System interpretability – Explore and translate each module and key forecasting algorithm to an application-based framework focused on actionable use-cases that helps the user to understand and review each outcome.
Accuracy improvement– Leverage and adjust all features to be tailored by business needs (disable, adapt, and develop new modules). Subsequently, adequate the parameters for each set of products.
Simulation & Monitorization – Use advance simulation to compare business KPIs between the old and the new process and carefully supervise forecasting accuracy over time.
To foster confidence and expertise within teams, using advanced simulation modules and explaining the concepts of each forecasting module were critical lever to success.
Consequently, an ambitious rollout plan was efficiently conducted with full support and comfort each operational team.
By bridging the gap between technical and business expertise was possible to adapt the new forecasting tool. This allows the system to be more responsive to business specificities and adjusted to the features of each product category.
Besides gains on processes and interpretability, the results showed more than 10p.p. improvement on accuracy and a more stable forecast compared with the initial design.
By: Bruno Batista
