This is a fictional composite inspired by recurring enterprise patterns.
At ten minutes to three every Tuesday, the corridor outside Room 4B filled with laptop bags, paper cups, and anxious product managers.
Inside sat the AI Review Committee of a large financial services enterprise. Outside, everyone carried the same thick template. David had filled his for a small internal assistant that sorted travel receipts. Beside him, Chris carried a model that would influence lending decisions.
“How long have you been waiting?” the lending lead asked.
“Three weeks,” David said.
He nodded sympathetically. He had been waiting five.
ProblemThe committee treated every idea as equally dangerous.
The committee was not careless or slow by nature. Its members were trying to protect customers and the company. But the process gave them fourteen files, one afternoon, and no simple way to separate a harmless internal experiment from a decision that could affect a person’s future.
So they asked everything about everything. Low-risk teams waited. High-risk teams received hurried attention. Evidence arrived as attachments nobody could find later. The queue grew, and people quietly began calling governance “the Tuesday tax.”
When every use case is urgent, important, and high risk, none of them is properly understood.
SolutionDavid asked for a road sign, not a shortcut.
During his review, Michael, the committee chair, asked if he wanted the process relaxed.
“No,” he replied. “I want to know which road I am on before I start driving.”
The room became quiet. Then Laura from security turned the whiteboard around.
Together they drew three lanes. The first covered contained experiments using approved tools and non-sensitive data. The second covered systems touching business operations or personal information. The third covered decisions with material impact, external users, or meaningful autonomy.
Each lane had a named owner, a short set of controls, and a clear approval path. Every team carried one living evidence card: purpose, data, model, risk tier, evaluation, human oversight, decisions, and exceptions. A change in purpose or autonomy sent the card back for reassessment.
Routine checks moved into delivery. The committee kept its time for judgment: unusual harms, difficult trade-offs, and the systems that genuinely deserved challenge.
OutcomeThe corridor became quiet.
A few Tuesdays later, David walked past Room 4B. There was no queue. His receipt assistant had entered the first lane with restricted data, an approved model, lightweight evaluation, and a named owner. Chris’s lending team was inside receiving the careful review its system deserved.
Michael stood at the door with a cup of coffee.
“No committee today?” David asked.
“Only one case,” he said. “At last, we have time to think about it.”
Governance had not become smaller. It had moved closer to the work. Teams could see the next responsible action, reviewers could focus on material risk, and leaders could trace why a decision had been made.
The practical takeaway
Classify early, apply controls in proportion to risk, keep evidence with the product, and reserve human review for decisions that need human judgment.