The Human-to-AI Leverage Ratio
How to calculate what an AI workforce actually displaces: hours not headcount, supervision in the denominator, capacity never reported as displacement, and no ratio of our own.
Version 0.1, draft, published 2026-08-11. How to calculate what an AI workforce actually displaces, published with no ratio of our own.
BvLogic publishes NO leverage ratio. It operates no AI workforce, has no customer estate, and has displaced no human work. When a ratio exists it will carry the scope, the period, the baseline and the humans still required, because a ratio without those four is a slogan.
Why the method comes before the number
This is the number every AI vendor wants to publish and almost none can defend. 'Equivalent to 78 FTE' is the most persuasive sentence available in this market and the easiest to manufacture, because every term in it is undefined: what counts as work, whose work, measured over what period, against a headcount that may never have existed.
So the method is published before any ratio is, and it is deliberately unflattering: applied honestly it produces smaller numbers than the ones in circulation, and the difference is the part worth having.
The error that produces most published ratios
Counting CAPACITY as if it were WORK DISPLACED. A system able to process ten thousand tickets an hour has not displaced ten thousand tickets an hour of human labour; it has displaced the tickets that were actually arriving, which is a much smaller number and the only honest one. Capacity is what a vendor demonstrates. Displacement is what a customer pays for, and the two are routinely reported as the same figure.
The calculation
human_hours_previously_spent / human_hours_now_spent, over the same scope and the same period
- Hours, not headcount. Headcount hides part-time, overtime and the person who did this alongside three other jobs.
- The SAME scope on both sides. A ratio comparing all of last year's work to this quarter's subset is arithmetic, not measurement.
- The denominator includes the humans the system created work for: approvals, exception handling, supervision and the engineer who maintains it. Leaving them out is the single most common inflation.
What a defensible ratio must carry
| Field | Why |
|---|---|
scope | Exactly which processes. A ratio for 'IT operations' means nothing; a ratio for 'password resets and stateless service restarts' can be checked. |
period | The window measured on both sides. Ratios computed over a busy month and compared to a quiet one are the oldest trick here. |
baseline_hours | Human hours actually spent BEFORE, from timesheets, ticket time or a time study. Not an estimate of what it 'should have' taken. |
baseline_source | Where the before number came from. A baseline nobody can trace is a number somebody chose. |
current_hours | Human hours spent now on the same scope, including approvals and exceptions. |
supervision_hours | Hours spent watching, maintaining and correcting the system. These are real and are almost never counted. |
displaced_work | What the system actually handled, in units that existed before: tickets closed, reports produced, incidents resolved. Not capacity. |
humans_still_required | Named roles that did not go away. A ratio implying nobody is needed is contradicted by the approval column in the audit log. |
confounders | What else changed. Headcount moves, a quieter quarter, a process simplification that would have reduced hours anyway. |
Rules
- Capacity is never reported as displacement. If the system could have handled more than arrived, the number is what arrived.
- Supervision hours are subtracted, not footnoted. A system requiring a full-time engineer to run has consumed a full-time engineer.
- A ratio is reported with its scope in the same sentence. '9.7x' alone is unpublishable; '9.7x on password resets and service restarts, Q3, against timesheet baseline' is a claim.
- No ratio is computed over a period shorter than one full business cycle for that process. Month-end work measured mid-month is not measured.
- Where a confounder cannot be separated, the ratio is reported as a RANGE and the confounder named. A point estimate that hides an unresolvable confounder is a false precision.
- A ratio is never extrapolated across processes. Leverage on password resets says nothing about leverage on incident triage, and multiplying one by an estate is how every inflated figure in this market was produced.
A worked example, and the size of the difference
Illustrative. A WORKED EXAMPLE with composed figures, describing no real organisation. It exists to show the arithmetic and, more usefully, how much smaller the honest number is.
| Scope | Password resets and stateless service restarts, second-line queue only |
| Baseline hours | 640 (Ticket time-tracking, second-line queue, same quarter last year) |
| Current hours | 95 |
| Supervision hours | 120 |
| Humans still required | One second-line engineer for exceptions; One platform engineer, 0.3 FTE, maintaining the automation |
| Confounders | Ticket volume rose 8% year on year; A password policy change reduced reset volume independently |
Naive: 640 / 95 = 6.7x
Honest: 640 / (95 + 120) = 3.0x
Reported as: 3.0x on password resets and stateless restarts, one quarter, against a ticket-time baseline. Range 2.6x to 3.0x because the password policy change cannot be separated.
The naive figure is more than twice the honest one, and the entire difference is the supervision column that most published ratios omit.
How this number gets inflated
- Reporting capacity as displacement, which inflates by whatever margin the system is oversized.
- Omitting supervision, approval and maintenance hours from the denominator. This alone commonly doubles a ratio.
- Baselining against a headcount that never existed: 'this would have needed 40 people' when the work was being done by four, badly, and nobody was going to hire 40.
- Comparing a busy before-period to a quiet after-period.
- Extrapolating one process across an estate, which converts a real 3x into an imaginary 30x.
- Counting work the system created for itself, such as tickets it opens and closes.
Questions to ask anyone quoting a ratio
- What is the scope, in processes a person could name?
- Where did the before number come from, and can you show me the source?
- Are supervision, approval and maintenance hours in the denominator?
- Is this displacement or capacity? What actually arrived in the period?
- Which humans are still required, by role?
- What else changed in the same window, and how was it separated?
- Has this been extrapolated from one process to an estate?
Status
Version 0.1, draft, and no ratio published. Applied honestly this method produces smaller numbers than the ones in circulation, which is the reason to publish it and the reason almost nobody does.
Licence. This specification is published under CC BY 4.0: implement it commercially, republish it, adapt it, with attribution and a note of what you changed. The conformance harness is MIT. Full terms and what is not covered: the licence page.
Everything here is free and stays free. There is no form in front of any document. If you want to know when new guides and templates go up, leave an email.
Roughly monthly. Unsubscribe in one click. We do not share your address, and we will not call you.