Zovda
A trend chart is designed to look convincing after the fact. By the time the line is steep, the decision that needed the number has already been made without it.
What a monitoring team needs is not a higher peak but an earlier bend.
Peaks are for reporting
A peak confirms something happened. It arrives after the attention, which is exactly when it stops being actionable.
The interesting moment is the week where the baseline moved by a small amount and stayed there. That is a change in slope, and it is boring enough that nobody puts it in a deck.
Baselines move on their own
Every account, topic and market has a rhythm: weekday and weekend, campaign cycles, seasonality. A raw count compares this Tuesday to a baseline that was never stable.
Model the baseline first, then look at what is left. Most "spikes" are the baseline doing its normal thing.
What a trend record holds
- The series, with the window it was computed on
- The baseline and how it was estimated
- The change, in absolute and relative terms
- The volume behind the change, so a move from three to six is not read as a surge
- The date the signal fired, not the date it was noticed
The fourth field is what stops a small number from becoming a strategy.
Backtest the alarms
Take last year's data, run the detector, and count how many alerts you would have sent and how many of them were worth reading. A detector that fires every day is a newsletter.
Zovda assembles these collections through Datuva, where each series keeps its window and its baseline. The data research collection shows how historical windows are built.
Tell us what you need to catch early. We reply within one business day.