Introduction
Every ROI figure in AI Governance is a count multiplied by what each counted session is worth — the What it is worth card on each agent you add to an outcome. It is the single most scrutinised value in the module, and the only one that is not derived from data.
Which means it has to be an agreed figure, not an estimate somebody typed.
What the field means
What it is worth offers two choices: A set amount — every counted session books the same figure — or An automation prices it. A set amount is best thought of as the cost avoided each time the outcome happens, weighed against the spend it takes.
Two parts of that do work. Cost avoided — not revenue generated, not value created in the abstract, but expenditure that did not occur because the agent did this. Each time it happens — per occurrence, not per month.
Arriving at a defensible number
Find the thing that used to happen — Before the agent, who did this and how long did it take? That is the cost being avoided.
Cost it at a loaded rate — Fully-loaded hourly cost times the time saved. A reviewer will accept a loaded rate; they will question an unloaded one.
Take only what the agent actually displaced — If a person still reviews the output, the agent avoided part of the task, not all of it. Claim the part.
Round down — A conservative figure that survives scrutiny is worth more than an ambitious one that does not.
"You will be asked where this number came from. The answer 'finance agreed it' ends the conversation. The answer 'we estimated it' starts one."
Who stands behind the number
There is no Owner field on an outcome; it shows the person who created it. So nothing in the product names who agreed the value — that has to come from you.
The value is the one input to ROI that a platform team should not be inventing on its own. Agree it with someone from the function that benefits, or from finance, before you enter it.
Changing the value later
Changing what a session is worth changes the figures the outcome reports from then on, and a number somebody already reported may not reconcile afterwards.
Treat a change as an event worth announcing rather than a quiet edit.
Sanity checks before you commit
Check | Why |
Multiply by a plausible monthly count | If the annualised figure looks absurd, the per-outcome value is wrong. |
Compare against the agent's spend | A value far below what a session costs means the outcome is not worth measuring. |
Compare against measurement cost | An AI-judged outcome adds spend per session. It must clear that too. |
Ask whoever owns the budget | If they would not defend it, do not publish it. |