Skip to main content
When a metric moves and you want to know why, the Why-agent runs a focused investigation: it digs through your releases, metadata changes, reviews, keyword movements, competitor moves, and past experiments, then returns ranked candidate causes with evidence — not a hand-wavy paragraph.

​
Where to use it

  • Dashboard → Co-founder → Why — pick the metric (revenue / downloads), a window, optionally sharpen the question (“why did US revenue drop after v2.4?”), and hit Run.
  • Chat — asking “why did revenue drop?” in Slack/Telegram/dashboard chat can route into the same analysis.
  • Signals — an anomaly signal can seed an RCA with its metric context attached.
  • API — see below.

​
What a report looks like

Every analysis produces a structured report:
  • Observation — what actually happened, quantified (“US revenue fell 34% between Jun 2–9”)
  • Candidate causes — each with a likelihood (very_likely / likely / possible / unlikely), evidence for, and evidence against. The agent argues both sides.
  • Top pick — the most likely cause with a confidence score and a concrete next step
  • Analyst notes — caveats and data gaps
If the agent can’t decide between two causes, it may register an open hypothesis — a signal that re-checks the question when more data lands.

​
API

Trigger an analysis (runs async, typically 30–90 seconds):
Then list / fetch reports:
Full request/response shapes: Why API reference.

​
Why it’s better than asking chat

Chat answers from the latest briefing plus a handful of tool calls. The Why-agent runs a dedicated pipeline with a bigger evidence budget, structured output, and hypothesis tracking — use it when the answer matters enough to want receipts.