A metric is useful when it changes what someone does. A dashboard full of numbers can still leave a team unable to decide whether a release helped.
Start from a decision
Before adding a chart, complete this sentence:
If this number changes, we will…
Possible endings include “investigate a regression”, “roll back a release”, or “invest in the faster workflow”.
Three kinds of measurements
| Type | Example | Question |
|---|---|---|
| Input | Number of published articles | How much work entered the system? |
| Behavior | Reading completion rate | What did people actually do? |
| Outcome | Return visits | Did the work create lasting value? |
Do not confuse activity with impact. More events can mean more usage, or merely more retries and confusion.
Include context
Always keep the release version, time window, and population near the metric. The same percentage can mean different things for new visitors and returning readers.
completion_rate = completed_reads / started_reads
window = 2026-08-01..2026-08-15
population = returning_visitorsThe goal is not to measure everything. It is to make the next decision less dependent on intuition alone.