Monitors

Your sales look steady this month. But is every store steady? Totals can hold flat while one region slides and another climbs, and checking every breakdown by hand, every week, isn’t a job anyone keeps up for long.

The idea

A KPI is one number for the whole business. A monitor watches that number per region, per store or per rep, and tells you which one moved.

What it watches, and when

Under the hood it’s a numeric column tracked over time, on a view or a source table. Start one from a KPI (your KPIs are listed under the Metrics tab) and it watches the KPI’s Measured from table. You can break the number down by up to two columns, like region, store or channel, and each value becomes a slice: West is one slice, East another. Up to 25 slices are kept, and the smallest of the rest are folded together.

Checks happen every time the watched view finishes rebuilding, and views rebuild after their sources sync. So it isn’t a nightly job, and it isn’t watching continuously. Check now on the monitor’s panel runs one on demand.

How it decides something moved

Each slice is compared with its own past, never with the other slices. West being a third the size of East every month isn’t news. West falling to a third of its own usual level is.

That’s why there are no thresholds to pick. The band a slice should stay inside comes from its own history, which needs at least 14 periods before the monitor judges anything. There’s no allowance for seasons.

Most of the time a monitor says nothing, and that’s normal. It speaks up when a slice leaves its band, and it holds back repeats, so a slice stuck at the same low level doesn’t alert again. Still broken isn’t news. Getting worse is. How much counts is your choice between Anything that leaves its band and Only changes worth acting on, for every slice at once.

Where you hear about it

When a slice really moves, a Something changed card lands in your Inbox. It names the monitor and the slice, says whether the number dropped or rose against what was expected, marks it worth acting on now or keep an eye on this, and lists other slices that moved too.

You can Acknowledge the card, or snooze just that slice for 7, 30 or 90 days with the button that names it (Snooze West, say) while the other slices keep alerting. Adjust sensitivity changes the whole monitor. Discuss starts a Librarian chat, which uses a credit.

A monitor can email its alerts, too. Everyone on its delivery list gets each one, and admins see alerts first in the nightly digest email. See Monitors and alerts.

What the card won’t tell you is why. A monitor knows which slice moved, not the cause, and that part is still yours to work out.

An example

Say you put a monitor on Curio’s Revenue by region view, broken down by region. Total revenue looks fine for the month, but West had a weak one.

The next time the view rebuilds, the monitor checks each region against its own history. East is bigger than West, as always, and that doesn’t matter: East is inside its band. West comes in well below its usual level. A Something changed card shows up saying revenue in West dropped, with the value it came in at, the value that was expected, and any other regions that moved.

Nobody had to ask. Nobody had to open a regional breakdown.

Where it lives in Querri

There’s no monitors settings page. You start a monitor in the Watching section of a KPI or view panel, with Monitor this and then Start watching, or by accepting a Worth watching card in the Inbox. Anyone who can edit the watched table can set one up, as long as the table has a date column and a numeric column. On a KPI, Monitor this appears only once the KPI has a Measured from table. View panels are covered in Table and column panels.

Click a monitor in that section to open its own panel, with a sparkline per slice, the biggest movers first, and its alert history. Alerts land as Something changed cards in the Inbox. On the Library’s Visualize canvas, a monitor is a small heartbeat mark beside the view row it watches, and it takes an alert color and a count while it’s alerting. See Monitors and alerts.

For data teams

Each slice is a time series at the watched table’s grain. A period is flagged when its modified z-score (based on the median absolute deviation, against a rolling median over 15 periods) reaches 3.5, and separate checks look for level shifts and trends. Nothing is judged before 14 periods, and there’s no seasonal model. Severity is watch or material. Repeats are held back by a cooldown, a same-level check and per-slice mutes. Slices are capped at 25, with the smallest folded into one.

Next steps

• How the Library learns
• KPIs