Asking why a number moved

Every analytics product on the market is good at telling you that a number changed. Almost none of them are good at telling you why, and that is the question people actually have.

The gap

A dashboard shows a drop. Someone notices it, opens three more dashboards, forms a theory, checks a deploy log, asks a colleague, and forty minutes later has an answer that was in the data the whole time. The work is real but almost none of it is thinking.

What an agent can do here

Watch the metrics between questions. When something moves, walk the obvious chain — segment, source, release, upstream dependency — before a person asks. Arrive with a ranked list of candidate causes and the evidence for each.

It will not always be right. It does not need to be. It needs to move the starting point from a blank dashboard to a shortlist.

What we are careful about

• A confident wrong answer is worse than no answer, so every claim carries its evidence.
• The agent never silently changes what it monitors.
• If it cannot explain a movement, it says so rather than reaching.

That last constraint costs us the most impressive-sounding demos and is the reason people trust the output.