What is Metric Investigation?

Metric Investigation helps you trace anomalies detected by Monte Carlo metric monitors back to their potential root causes — faster and with more context. With this feature, you can sample and segment rows related to a metric alert, giving you immediate visibility into the underlying data that triggered an alert.

Metric Investigation provides a lightweight root cause analysis workflow built directly into the alert page.

Why Use Metric Investigation?

When a metric monitor triggers an alert, teams often ask:

What is driving the change in this metric?
Are certain values or dimensions contributing to the anomaly?
Are there unexpected outliers or data quality issues in the rows behind this result?

Metric Investigation helps answer these questions by making it easy to:

View sample rows related to a metric alert.
Segment rows by field to quickly surface correlations.
Drill into anomalous vs. non-anomalous behavior.
Compare across time periods or segments.
Copy and refine the SQL query for deeper investigation.

This reduces the time from detection to understanding, helping data engineers and analysts diagnose and respond to data issues more effectively.

Overview

When viewing a metric monitor alert, you will see an Investigate rows icon on each event and under the Investigate button. Clicking this opens the investigation drawer, where you can explore:

1. Sampling and Segmentation

Sample Rows: View a sample of rows for the alerting measurement.
Example: Rows where mp_processing_time_ms < 0 triggered a Negative (%) alert.
Segment by Field: Group rows by any low-cardinality field on the table to spot correlations.
Example: Group by status to see counts of paid, refunded, etc.
SQL Query: The exact SQL used to retrieve the displayed data is visible and can be copied for further analysis.
Sample Rows Table or Bar Chart: Depending on the type of investigation:
- Full table of rows (useful for outlier review and deep dive).
- Bar chart of segment distribution (useful for correlation and comparison).

2. Investigation Modes

You can switch between multiple investigation modes:

Sample: View anomalous rows for the alert condition.
Segment: Segment anomalous rows by field (e.g. client, region, status).
- Segmenting works best on fields with low to moderate cardinality, as it's limited to the first 1,000 segments.

3. Supported Metrics and Fields

You can see the full list of Available Metrics in metric monitors in the linked documentation. All metrics are supported for investigation, but the options to filter rows by may slightly vary per metric.

Metric Type	Supported?
Pipeline metrics (Row count, Relative row count, Time since last row count change)	1. Only "all rows" since there is no differentiation between which caused an alert.
Field metrics (distribution or counts, e.g. Mean, Median)	1. All rows from the time of the alert 2. Rows where the field is above the metric 3. Rows where the field is below the metric
Field metrics (conditions, e.g. Null %, Unique %)	1. All rows from the time of the alert 2. Rows where the field meets the metric condition 3. Rows where the field does not meet the metric condition
Custom metrics	1. All rows from the time of the alert 2. Rows where the field meets the metric condition 3. Rows where the field does not meet the metric condition

Example Investigations

Example 1: Sample metric underlying rows

Alert: Negative (%) on mp_processing_time_ms
Investigation: Sample rows where value < 0
Segment: mp_browser and mp_event_name

Example 2: Segmentation

Alert: Sum of invoice_amount outside expected range
Investigation: Segment rows by status to understand row count distribution by type (paid, refunded, unpaid, etc).

Notes and Tips

SQL-first: Everything shown is backed by a SQL query you can copy and modify.
Anomalous rows first: Investigation focuses on showing you the rows that drove the alert.
Segment smartly: For tables with many columns, choose low-cardinality fields for the clearest results.

Learn More

For background on how metric monitors work, see the Metric Monitors Overview.