The Judgment Multiple
The Judgment Multiple (IgnatiusTheYoungerAI, 2026) is a metric for pricing human verification work against the AI errors it prevents.
Definition
Judgment Multiple = annualized exposure ÷ annualized cost of the human control.
Expressed as a ratio. A Judgment Multiple of 18x means that for every dollar spent on a defined human verification step, roughly eighteen dollars of expected loss is addressed.
Why it exists
Organizations adopting AI measure one side of a ledger. Time saved per task, volume of output, adoption rate, headcount avoided — all of these have dashboards and monthly reporting. The other side has nothing: the cost of errors that get through, the cost of verification labor, who absorbs the consequence, and decisions made on wrong information.
When one side of a ledger is instrumented and the other is not, the uninstrumented side loses every argument it enters — not because it is wrong, but because it cannot produce evidence in the format the conversation requires. The Judgment Multiple is an attempt to instrument the second side using arithmetic a finance team already accepts.
How to calculate it
- Observe. Log AI-assisted outputs for 30 to 60 days. Record what was verified and what was found. This produces a measured error rate over a stated window.
- Model the exposure. Annualized output volume × observed error rate × the share of those errors the control actually catches × estimated cost per uncaught error. State every assumption.
- Measure the control. Time the control with a timer, not an estimate. Multiply annualized control hours by your fully loaded hourly rate.
- Divide. Present a range, not a point estimate, and label the modeled components as modeled.
The catch-rate term
A Judgment Multiple calculated without a catch-rate discount assumes the control intercepts 100% of the errors it is aimed at. No control does. Multiplying expected errors by a defensible catch rate — 70% is a reasonable starting point — produces a lower ratio that survives scrutiny. A ratio that cannot survive scrutiny is worth less than no ratio at all.
What it is not
It is not a measured loss figure. It is not an ROI claim about an AI tool. It is not a substitute for an incident record. It is a modeled ratio built from stated assumptions, and it should always be presented as one.
Origin
The Judgment Multiple was introduced in AI "Keep Your Career" Bible: The 24 documented failures in AI that prove why you're still necessary — and the 90-day plan to prove it to everyone else, by IgnatiusTheYoungerAI, first edition, 2026. The book works the ratio for all 24 documented failure modes and shows the assumptions behind each one.