Skip to main content

Insights Service: Answerable Question Patterns

The insights service reasons over pre-computed statistical insights for governed metrics — it does not query raw data rows.

Can answer​

Question patternInsight type(s) involved
"What's happening with [metric]?" / "Summarize [metric]"Current Trend, Unexpected Values
"Why did [metric] go up/down?" / "What's driving [metric]?"Top Drivers, Top Detractors, Top Contributors
"What changed most?" / "What's unusual?" / "Any anomalies?"Unexpected Values, Trend Change Alert
"Which [dimension] is highest/lowest?"Top Contributors, Bottom Contributors
"Which [dimension] is concentrated / dominating?"Concentrated Contribution Alert
"Is there a trend?" / "What's the trend direction?"Current Trend, Trend Change Alert
"Are we on track to hit goal?" / "Pace to goal?"Pace to Goal, Forecast
"What's the forecast for [metric]?"Forecast
"How does [metric] compare to a threshold or goal?"Goal and Threshold Breakdown
"Which records or transactions are outliers?"Record-Level Outliers
"Is [metric A] correlated with [metric B]?"Correlated Metrics
"Tell me about [metric]" / "Give me an overview" / "What should I focus on?"All of the above combined

Cannot answer​

  • Questions requiring custom SQL or arbitrary data slices not modeled as metrics
  • Questions about data with no corresponding metric definition on the site
  • Questions about workbook layout, viz formatting, or content catalog navigation
  • Real-time streaming data or data below the metric's configured granularity

Edge cases (handle, don't silently fail)​

  • Metric has no goals — pace-to-goal / on-track / threshold questions can't be answered for it. Say the metric has no goal configured and offer trend/forecast instead; omit goals from the payload.
  • Cross-datasource multi-metric — the brief accepts metrics from different datasources but produces weaker, less-comparable results. Prefer grouping by datasource and making one call per datasource; if the user insists on comparing across, proceed but flag the reduced reliability.
  • Dimension asked about isn't in allowed_dimensions — the service can't break down by it. Answer at the metric level and note the requested breakdown isn't available for this metric.
  • Granularity below the metric's configured level (e.g. hourly on a daily metric) — out of scope; answer at the metric's granularity and say so.