ThresholdIQ: Rebuilt the Engine From the Ground Up

A few weeks ago I launched ThresholdIQ to help teams monitor spreadsheet data without needing heavy BI tools like Microsoft Power BI or Tableau.

The first version used simple threshold rules: upload a file, set limits, and receive alerts when values crossed them.

It worked — but one issue kept appearing.

Setting thresholds is harder than it sounds.

Seasonal revenue, slowly drifting metrics, and correlated data often break rule-based systems. So instead of improving the rule engine, I removed it.

ThresholdIQ v2 now uses 9 concurrent anomaly detection methods to detect unusual patterns automatically.

When you press Detect, the engine runs multiple analyses directly in your browser:

multi-window Z-score baselines

EWMA deviation detection

SARIMA seasonal modeling

Isolation Forest outlier detection

correlation deviation checks

DBSCAN clustering

Additional detectors catch seasonal hour/day effects, slow trend drift, and stuck or zero-value signals.

Each method contributes to a weighted score that determines alert severity: Warning, Critical, or Emergency.

Why everything runs in the browser

All analysis happens locally — your data never leaves your machine. No server processing, no external APIs, no third-party calls.

The ML libraries are bundled directly into the app so everything runs offline.

Tech stack: Vanilla JS, Web Workers, Chart.js.

If you regularly monitor spreadsheet data, I’d love feedback.

Try it here: https://thresholdiq.app