Why this number, and not the others
Every AI vendor reports hours saved. Hours saved are real and they are not money. A company that saves 20 hours a week and doesn't do anything with them has the same payroll and the same revenue it had before. Adoption metrics are worse: seats purchased, prompts run, workflows built. None of it touches the income statement.
Revenue per employee does. It is revenue divided by full-time-equivalent people, and it is the number that goes up when AI is absorbing work that would otherwise have been hired, and stays flat when a company is collecting tools. It also happens to be the number acquirers and lenders look at first, and the one that makes "we use AI" a sentence with evidence behind it.
How to compute it without fooling yourself
The definition is simple and the edges are where people cheat, usually by accident.
- Revenue: trailing twelve months, so one good quarter doesn't flatter you. If the business is seasonal, compare the same month year over year as well.
- People: full-time equivalents, including the owner, including part-timers at their fraction, including long-term contractors who do employee work. Excluding contractors is the most common way to make the number lie.
- Timing: compute it monthly, using the headcount on the last day of the month. Write down the hires you avoided in the same spreadsheet, with the role and the loaded cost, so the number has a story.
- Acquisitions: an acquired company brings revenue and people at the same time. Track the combined entity from the close date and don't compare it to the pre-close number.
Worked example: $8M trailing revenue, 40 FTEs, revenue per employee $200,000. The plan says four hires this year. Grow 15% and hire the four: $9.2M over 44 people, $209,000. Grow 15% and absorb three of the four with systems: $9.2M over 41, $224,000, plus roughly $225,000 of loaded payroll that stayed in the business. Same growth, different company. The calculator on our audit page runs this with your numbers.
The three things that move it
Only three, which is why the number is useful.
- Hires avoided. The admin, the coordinator, the second SDR, the ops manager: roles whose work is mostly rules-based and now runs as a system. This is the fastest mover and the one most companies never count because the hire simply didn't happen.
- Hours returned to people you already pay. The senior person who stops assembling reports and starts selling again. Hours only count here if they turn into revenue or into a hire avoided, which is why the report tracks what the hours were used for.
- Revenue the systems produced. Meetings booked by the pipeline, inbound from the pages the AI engines now cite, quotes that closed because the follow-up went out in an hour instead of a week. Attribute it honestly, by the lead source the CRM recorded, not by vibes.
Everything else an AI program does rolls up into one of those three or doesn't matter.
The one-page monthly report
This is the page we send clients. It replaces the dashboard nobody reads.
| Line | This month | Last month | Note |
|---|---|---|---|
| Revenue, trailing 12 months | $8.4M | $8.1M | From the accounting system, not the CRM |
| Full-time equivalents | 40.5 | 40.5 | Owner included; two part-timers at 0.5; one contractor at 0.5 |
| Revenue per employee | $207,400 | $200,000 | The number |
| Hires avoided (cumulative) | 2 | 1 | Intake coordinator (Mar), reporting analyst (Jun), loaded $150K/yr |
| Hours returned per week | 31 | 24 | Of which 18 redeployed to client work, 13 to the owner |
| Systems-sourced revenue, trailing 3 months | $310K | $260K | By CRM source: outbound 11 deals, AI-cited pages 4, follow-up recovery 6 |
| What changed this month | Quote follow-up sequence rewritten; after-hours intake live; two pages now cited by Perplexity |
Two rules make the report trustworthy. The revenue line comes from accounting, never from the CRM. And "hires avoided" only counts a role that was budgeted, named and not filled, with the date; a vague "we would have needed someone" doesn't count.
What not to do with it
The number can be gamed by cutting people, and a company that games it will find out why Klarna hired customer-service people back after its very public AI cuts: service quality is revenue, eventually. The right way to move revenue per employee is to grow without the next hires and let attrition and redeployment do the rest. The wrong way is layoffs with a spreadsheet.
It is also a trailing number. Systems take a month to build and a quarter to show up in revenue, so the first report after a build will show hours returned and a hire avoided before it shows revenue. That is the expected shape. The companies that get this right keep the page going for a year and read the trend, not the month.
If you'd rather have it mapped for you: the free Revenue-per-Employee Audit takes the hires you're planning, shows what a system absorbs from each, and prices the 90-day build. Written, within three business days.