KPI Anomaly Alert System

Overview

This project demonstrates an automated system for detecting abnormal fluctuations in key business KPIs.

The system monitors daily performance metrics and identifies anomalies using a robust statistical method based on Median and Median Absolute Deviation (MAD).

Problem

Business teams often detect KPI drops days or weeks later during periodic reporting.

This delay can result in:

• product issues remaining unnoticed
• tracking bugs affecting reporting
• delayed response to algorithm or UI changes

The goal of this system is to provide same-day visibility into KPI anomalies.

System Architecture

Data Warehouse → Power BI Model → Anomaly Detection → Power BI Service → Power Automate → Email Alerts

Detection Logic

Step 1 — Fetch last 5 same-weekday KPI values
Step 2 — Baseline = Median(values)
Step 3 — MAD = Median(|value − baseline|)
Step 4 — Spread = MAD × 1.4826
Step 5 — Upper = Baseline + 2.5 × Spread
Step 6 — Lower = Baseline − 2.5 × Spread

If the KPI value falls outside the range → anomaly detected.

Alert Classification

Critical ≥ 35%
Major ≥ 25%
Minor ≥ 15%
No Alert < 15%

Tech Stack

SQL
Power BI
DAX
Power Automate
SMTP Automation

Impact

• Automated KPI monitoring
• Same-day anomaly detection
• Faster issue identification
• Reduced manual dashboard monitoring