The bank's AI that never decided

A global bank approved an AI system to automate routine credit decisions in late 2022. It processed applications faster, reduced errors, and promised cost savings. Six months later most files still flowed through human review.

Authority was never granted

The system worked and had formal approval. But it never received the right to make decisions within agreed-upon boundaries with the organization's full backing. Exceptions multiplied and temporary safeguards became permanent.

The pattern repeats across industries

A global manufacturer rolls out AI demand forecasting but planners treat it as a second opinion. A healthcare network introduces AI triage that reduces wait times, yet clinicians manually review most recommendations.

The real problem: governance lag

Organizations adopt systems that make better decisions but preserve human authority structures that prevent those systems' decisions from being accepted. Structures for accountability and legitimacy have not evolved to support AI.

ERP systems faced the same problem

When ERP systems like Oracle and SAP were introduced in the nineteen nineties, they promised centralized visibility and optimization. Nestlé USA's divisions had operated autonomously, each accountable for its own production and costs.

Nestlé treated it as technology, not redesign

The ERP system required divisions to surrender discretion to a common process. Employees accountable for local outcomes had good reason to keep making decisions themselves. The system produced one forecast; accountability rewarded local overrides.

Credit scoring showed the same pattern

Automated credit scoring models codified risk faster and more reliably than humans. Studies of auto lending show algorithmic underwriting produced higher loan profits and lower default rates than human underwriting.

But loan officers still overrode the models

Override cultures abounded. The biggest performance gap was where loan officers had the strongest incentives to preserve their own discretion. Loan officers manipulated inputs to push borderline applications over approval thresholds.

Airlines got it right with revenue management

Revenue management systems achieved lasting authority because airlines reassigned accountability. They clarified when human intervention was appropriate, changed performance evaluations, and made adhering to the system professionally defensible.

Authority shifts when governance changes

Authority failed to move when decision logic migrated faster than accountability structures could evolve. When governance changed, authority could shift. The pattern is consistent across industries and technologies.

The real reason for overrides

Authority fails to move when people remain personally liable for the system's outcomes. Employees override AI systems not as resistance but as self-protection. No training or communication can overcome this gap.

“If the system makes the call and something goes wrong, it is my name on the report. If I make the call, at least I can explain it.”

A dispatcher's honest explanation

A dispatcher at a global logistics firm explained the logic: if the system makes the call and something goes wrong, it is his name on the report.

Agentic AI raises the stakes

Agentic systems do not recommend; they act. They book, approve, route, contract, and escalate within authorized parameters. When an agent acts and the outcome is bad, accountability becomes genuinely difficult and ambiguous.

Four questions leaders must answer

Before scaling an AI system, leaders must explicitly answer: Who loses discretion? Who gains authority and over which decisions? Who explains outcomes when the system is followed? Who absorbs transition risk?

Amazon's successful authority transfer

Amazon's Hands off the Wheel initiative, begun in two thousand twelve as Project Yoda, embedded forecasting and pricing predictions in workflow tools.Business teams shaped data and parameters, but routine overrides were discouraged or required approval.

Accountability moved with authority at Amazon

Vendor managers shifted from individual buying decisions to auditing inputs, exceptions, and outcomes. The transfer stuck because accountability moved with authority. When results missed expectations, the question was which input or system design failed.

Redesign roles before scaling systems

Technology and organizational architecture must move together. Leaders must redesign roles so human judgment shifts upward into oversight, boundary setting, and system refinement. Employees need tools to explain decisions they did not personally make.

The measure of real adoption

Ask: If the system were turned off tomorrow, would decisions actually change?Or would people continue making decisions as they always have, only without the interface open?If decisions would not change, authority has not moved.