Board brief · Judgment Capital
The judgment capital problem
What a board is accountable for when AI does the first pass
Every institution runs on a stock of judgment that it built over decades and has never measured. As AI takes over the routine work through which professionals learned their craft, that stock runs down while every reported number improves.
1The asset
Judgment behaves like capital. It accumulates slowly, it is spent quickly, and it appears on no statement, yet the institution runs on it every day. Its size becomes known on the day someone has to override the machine.
The capital can also be issued. With the right cues and fast, clean feedback, judgment forms faster than the folklore of mastery suggests. Some institutions are building that bench today.
2Why the board acts before the proof
The decisive study would compare professionals formed the old way with professionals formed alongside the machine, over a decade. The first group shrinks every year, so the evidence will arrive after the damage. Boards already act on risks of this shape: key-person risk is written down without a trial. What discipline asks is that each entry be falsifiable and carry the condition under which it comes off the register.
3Six ways the asset drains
One stock, six mechanisms, each with an owner the board can name. Succession (is the pathway still open?) belongs to the chief HR officer. Decision quality (is the expert eye still there?) belongs to the chief risk officer.
| Mode | What the board would see | Owner |
|---|---|---|
| Pipeline collapse | Juniors who never arrive, and specialists with no one forming behind them | CHRO |
| Competence drift | Override rates sliding toward zero as reviewers converge on the machine's view | CRO |
| Inquiry groupthink | Every seat briefed to the same depth from the same fluent source, and a room that stops producing alternatives | CRO |
| Authorship erosion | A growing share of output that nobody on the payroll could rebuild with the tools down | CRO |
| Agency erosion | A correct judgment that forms in someone and is stopped by process before it reaches the decision | CRO |
| Detection blindness | Watchers who can no longer see what the machine missed; it arrives as a bill | CRO + CHRO |
4Five questions for the next meeting
Each is answerable by the next meeting if the institution is sound. Every good answer is a name, a number, a date or a record.
Who overrode the machine?
How many times last quarter did our people overrule our most consequential automated system, and what makes that number healthy?
A good answer
A number, its trend, and a floor set in advance.
The half-answer
Praise for a falling number, or the machine's accuracy.
Who decided what to protect?
In our most automated process, which steps still build the judgment of the people who run it, and who chose them?
A good answer
A named owner and a step-by-step map of where judgment forms.
The half-answer
Every step was assessed for automation potential.
Tools down, who catches it?
If our most senior credit, claims or audit judgment left on Friday, who catches what she catches, and when did we last test it with the tools off?
A good answer
A name and a date.
The half-answer
A succession plan for her chair, with nothing about her eye.
What did nobody see?
What share of the machine's decisions last quarter did no human look at, and who samples that share, how often, against what standard?
A good answer
A percentage, a sampling owner and a schedule.
The half-answer
The model is continuously monitored.
Where is it written?
Where are the first four answers recorded, who verified them, and would the record satisfy an auditor?
A good answer
A document with a verifier's name and a review date.
The half-answer
Minutes recording that assurances were given.
The motion
That management answer these five questions in writing within sixty days; that a named person verify the answers; and that the answers return to this committee as an agenda item.