How we measure an agent fleet in full-time equivalents

Numbers like "our AI does the work of 2,500 people" deserve suspicion. Most of the time nobody can explain them. So here is our methodology, in full, so you can challenge it — or apply it to your own operations.

The unit: hours, not magic

Every agent we deploy is instrumented. Each action it completes — an email processed, a ticket resolved, a follow-up sent, a report written — is logged with a time equivalent: the average time the same task took a human before automation, measured during week one of the project.

We sum those hours across the fleet over a rolling month, then divide by the legal working time of a full-time employee (151.67 hours per month in France). That quotient is the fleet’s full-time-equivalent workforce. No projections, no potential — only completed, logged actions.

What the number does not say

• →It does not mean 2,500 jobs were replaced — most of these hours were simply not being done, or were done late, at night, or never.
• →It counts execution time, not the judgment humans add on escalations.
• →It moves: agents get added, improved, and sometimes retired.

Why we publish it anyway

Because it is the honest way to answer the question every executive asks: "how much work does this actually do?" An adoption dashboard means nothing to a board. A workforce equivalent does. If you deploy agents, instrument them from day one — the number will do your quarterly reporting for you.