Organizations rely heavily on KPIs to guide decisions, but identifying when a metric is truly behaving abnormally remains difficult in practice. Simple rules and fixed thresholds often break down when data includes seasonality, trends, noise, and structural changes, which are common in real-world business metrics.
In this hands-on tutorial, participants will learn how to design and implement practical anomaly detection workflows for KPI data using Python. The tutorial focuses on building intuition first, then translating that understanding into working code that can be adapted to different domains.
We will begin by defining what anomalies are, what they are not, and why KPI data requires different treatment than static datasets. Participants will explore statistical baselines, time-series approaches, and machine learning methods for anomaly detection, with a strong emphasis on interpretability and evaluation. The tutorial includes guided exercises using Python libraries such as pandas, scikit-learn, and PyOD.
Real-world People Analytics examples, such as employee turnover, engagement trends, and role movement, will be used throughout the tutorial to illustrate common challenges. These examples are chosen because they are intuitive and noisy, but the techniques apply directly to other business KPIs in finance, operations, product, or marketing.
The tutorial also includes a focused section on when and how generative models can support anomaly detection workflows, particularly for summarizing patterns and supporting human interpretation, while clearly discussing their limitations.
By the end of the session, participants will have a solid mental model for anomaly detection, hands-on experience implementing multiple approaches in Python, and a practical framework for choosing the right method for their own data.