Watch
Tell me when something actually changes.
Watch learns what normal looks like for your number and speaks only when a day is really off.
Revenue watch
Example store · every morning at 07:00
Revenue is below its normal range for a Saturday. Every Saturday in the last six weeks was higher; weekdays at this level are ordinary, which is why a fixed alert would have missed it.
Not a threshold
An alert rule is a guess about normal, made once and never revisited.
The same six weeks of one shop, judged by a fixed line and by Watch.
How it decides
Normal, measured the way a statistician would.
The maths, with worked numbers, is in rolling baselines, explained.
A typical level
The median, so one record day cannot redefine normal.
A typical spread
How far ordinary days wander. It sets the band's width.
The weekly pattern
Each weekday judged against its own normal.
SMTWTFSEnough history first
No verdict until 14 readings. Until then it says it is learning.
Sensitivity you set
Low, normal or high: how far out a day must be.
Consistent every run
The same statistics every time. The AI never does the maths.
The verdict
Not just that it moved. When, and what moved it.
How far off
In typical spreads.
How long
First odd day, or third in a row.
Since when
The day a new level began.
What drove it
Each segment against its own normal.
- Level shift
- ▼ since 1 Sept
- New level
- $41,640 (was $48,001)
- Size
- 4.4σ
What drove it
Each channel against its own usual level.
Quiet by design
It emails when something changes, then stops.
On change
Not every morning it stays off.
Silent on normal days
No email means nothing unusual.
Escalations too
A change that turns serious writes again.
Nothing lost
Every reading is kept in History.
$7,252
Below its normal range for a Saturday.
Put a Watch on the number you check every morning.
A source with a date column and a few weeks of history is enough to judge from the first run.