Quantify Analytics Dashboard

Overview

Quantify's early customers loved the product but kept asking the same question: 'is this number good or bad?' We built a self-serve analytics dashboard with a machine learning layer underneath that automatically flags meaningful anomalies, so users didn't need a data team to make sense of their own metrics.

The challenge

Off-the-shelf anomaly detection tends to either flood users with false positives or miss real issues entirely, especially on the kind of noisy, seasonal data Quantify's customers had. The model needed to be accurate enough to trust, and explainable enough that a non-technical user could understand why it flagged something.

Our approach

• Trained a custom anomaly detection model on Quantify's historical customer data, tuned specifically to reduce false positives on seasonal and weekly patterns.
• Built an explainability layer that shows users why a data point was flagged, not just that it was — comparing it against expected ranges in plain language.
• Designed a self-serve dashboard builder so customers could create custom views without engineering involvement.
• Set up a BigQuery-based data pipeline that handles high-cardinality event data without slowing down dashboard load times.