Metric Logic Tree: Automated Anomaly Analysis for Business Metrics

All businesses face simple yet hard‑to‑locate questions such as why a metric rises or falls. Traditional analysis using multi‑dimensional queries or dashboards often takes a day and is hard to transfer when analysts change.

The Metric Logic Tree abstracts a best‑practice smart anomaly analysis product. It combines fixed analysis methods with flexible data sources (Kylin, MySQL, Elasticsearch, Druid) to automatically generate anomaly reports for various users, directly providing conclusions and reducing decision cycles.

Challenges include numerous base metrics and dimensions from heterogeneous sources, the need to support multiple anomaly‑detection algorithms, and custom‑computed metrics.

The solution consists of three layers:

Metric calculation – resolves heterogeneous sources and custom formulas.

Analysis algorithms – currently a waterfall iterative method and a Gini‑coefficient based contribution analysis.

Computation service – a master‑worker architecture that splits tasks, runs parallel queries, and merges results.

The waterfall method identifies the most influential metric in a chain, while the Gini coefficient quantifies each dimension’s contribution to a metric’s fluctuation. Parallel computation handles the large volume of fine‑grained queries generated by user‑defined metric sequences.

In practice, the Metric Logic Tree has been applied across Meituan‑Dianping’s hotel‑travel lines, receiving positive feedback, though the product is still in its early stage and will continue to evolve.