Skip to main content

Metrics

Using Metrics
Define a metric, scope it to a Workflow, and read its results. Read the transcript

A metric defines an outcome to follow over time. Create one for a specific Workflow, then read its definition and chart with the date range in mind.

Before You Begin​

  • Have permission to create and view Metrics.
  • Choose the Workflow and outcome you want to measure.

Define a Metric​

  1. Open Metrics. The cards summarize measurements over the last thirty days.
  2. Choose New metric. Start with an AI Score, CEL Expression, or Member Count, or select an example such as Escalations. For CEL criteria, the Developer CEL reference lists variables available in metric context.
  3. If you start from an example, review the expression it fills in. Name the metric, choose the Workflow it measures, then create it.
  4. Open the metric to review its definition and chart.

An empty chart means there are no readings in the selected range. It does not mean the measured value is zero.

Choose What to Measure​

MetricConfigureReading
AI ScoreA grading question and passing thresholdAn AI score for each evaluated Chat
CEL ExpressionA rule over the supported metric contextThe expression's result for each evaluated Chat
Member CountA saved Member FilterThe number of matching Members, sampled daily

The metric type stays fixed after creation. Create a separate Metric if you want to measure a different kind of outcome. Examples such as Escalations, Resolution rate, and Buying intent fill in a starting definition; review and edit it before saving.

Read the Results​

Open a Metric and choose the date range. Check its definition, denominator, and attribution window before comparing readings. Open a reading's linked Chat when you need its context.

DisplayHow to interpret it
RateA 0–1 ratio displayed as a percentage
ScoreA 0–1 score displayed to two decimal places
CountA total: passing AI readings, or the sum of expression values
Member CountThe latest cohort size, not the sum of daily snapshots
CustomThe numeric value without special formatting

Rate, Score, and Custom Chat metrics show an average for the selected window; Count metrics show a total.

For Chat metrics, Observations uses each reading. Unique Members keeps each Member's highest reading within the window before calculating the result; readings without a Member remain separate observations. For example, three escalations among nine observations give a rate of 33.3%. Check the Readings number if your expected denominator differs.

AI count metrics count readings that meet the passing threshold used when they were graded. Changing the threshold affects future grades; existing readings keep their recorded result unless recomputed. Expression counts add numeric results. Daily or weekly chart buckets use the same aggregation as the headline.

Set a Target​

Add a target value and due date to the Metric. Review progress on its card, the target line in its chart, or the Metric targets dashboard widget.

Review Grader Notes and Export​

AI grader notes can quote Chat content. Treat them as patient data when the Chat contains it. Use Show grader notes in the readings table to inspect them; viewing readings and exports requires analytics:read.

Choose Export and the format you need:

  • Aggregate exports one numeric row per time bucket, without grader notes.
  • Raw readings exports each observation, including grader notes, even when the chart uses Unique Members. This export is audit-logged; handle it through your approved process for patient data.

Pause or Archive a Metric​

Turn off Compute this metric to stop new readings while keeping its history. Archive it to hide it from the list. Use Show archived to find it again.

Permissions​

Viewing definitions requires workspace:read; creating or changing them requires workspace:write. Readings, charts, and exports require analytics:read.