Introducing Anomaly Watch: your metrics, monitored 24/7

Most dashboards tell you what happened. Anomaly Watch tells you when something unusual happened — and why it probably did.

How it works

• Every numeric metric is aggregated into a daily series.
• For each day we compute a rolling median and MAD (median absolute deviation) over the previous 14 days (configurable from 7 to 30), in log space.
• If today’s robust z-score is beyond 3σ (also configurable), it’s flagged. Beyond 5σ it’s marked critical.

Why robust, log-space z-scores?

• Rolling baselines adapt to growth and seasonality without a heavy model — a Diwali spike is compared against the fortnight before it, not last summer.
• Median + MAD aren’t dragged around by the very outliers we’re trying to catch, unlike mean and standard deviation.
• Log space treats a 50% drop the same whether you do ₹5,000 or ₹5 lakh a day.

Root-cause hints

When a day is flagged, DataOps breaks it down by your primary dimension (city, channel, category…) and tells you which value drove it. That turns “revenue dropped” into “revenue dropped, and 60% of the day came from Ahmedabad”.

Try it

Load the Quick-Commerce Orders sample — it contains a Republic Day spike and a three-day outage dip. Open Anomaly Watch and move the sensitivity slider to see detection change live.