In April, our median cycle time was 2.8 days. In May, it was 2.17 days. We got these figures by connecting Jira tickets to a Grafana dashboard. Does that prove AI saved us time? No. Two months is too short, and several factors could explain the difference. But at least we are starting to see what is happening in our development flow.

Our internal dashboard (Grafana): median cycle time, month by month.
What we measure
A task's cycle time is the time between the moment it starts and the moment it ends. Each month, we take the median over the tasks completed: half took less time, the other half more. A median holds up better than an average against a few tickets that stayed blocked for weeks.
We also track actual processing time, time spent in review and the time a task stayed blocked. Together, these indicators show where work moves forward and where it waits.
A month-to-month comparison is only valid on the same scope (here, a given project) and with the same definition. Change the rule without saying so, and the curve compares different things. Even then, one month's ticket mix is not necessarily like the next.
How the figures are produced
Every night, our tool retrieves the Jira tickets, reconstructs the stages they went through and calculates a few indicators. The results are stored in PostgreSQL, then displayed in Grafana. When a data point is missing, we do not invent a value to fill in the chart.
The calculation is deterministic: the same month, recalculated, gives the same result. Ingestion runs as a full or incremental sync, and the dashboards read the data in read-only mode, through an API protected by a token.
What two months do not let us conclude
The drop is real over the period, but the period is short. It proves neither a trend nor the effect of AI. A shorter cycle time can come from tooling, organization, the nature of the tasks or simply a more favorable month.
These indicators describe a process, not people. We do not use them to rate anyone: that would be the best way to distort the measurements.
We will keep measuring, with the same definition and over several months. Only then will we be able to tell a good month from a lasting improvement.
