All hell broke loose: AI made the wrong decision for HR
It started on a Tuesday morning. The HR team at a mid-sized logistics company had been using an AI-powered performance management tool for eight months. The tool tracked output, flagged underperformers, and generated quarterly ratings — and until that point, senior leadership was very happy with the results.
Then the quarterly ratings landed in employees' inboxes. The AI had rated forty-three employees as low performers — including three of the most respected team leads, two senior engineers who had just delivered a critical system upgrade, and an HR professional who had personally trained the team that implemented the AI tool.
The ratings were completely wrong. It was later discovered that the AI had weighted a single metric — response time on a ticketing system — far above every other measure of performance. Employees managing complex projects that naturally delayed ticket response time were flagged as underperformers. Employees who had done nothing but close low-priority tickets quickly were rated as stars.
"The AI tool did not cause this crisis. The organisation did. And until HR teams understand that distinction clearly, they will keep deploying AI in ways that will lead to disasters."
— Team peopleHum, peopleHum BlogWhy AI Gets HR Decisions Wrong
AI does not make mistakes the way humans do. It does not have a bad day, miss a detail, or let personal feelings cloud its judgment. But it makes a different kind of error. AI makes decisions based on the data it was trained on and the metrics it was told to optimise. When those metrics are incomplete or poorly chosen, the AI produces systematically flawed outputs.
In the logistics company's case, the AI was optimising for speed. Nobody had told it that speed was only one of many things the organisation cared about. Nobody had reviewed whether the metric it was weighting most heavily was the right one for the roles being assessed. Nobody had tested the outputs before they were sent to employees. The system ran. That is not an AI failure. That is a governance failure — and it is far more common than most organisations realise.
The Consequences When AI Gets an HR Decision Wrong
When AI makes the wrong decision in HR, the consequences are immediate, visible, and expensive. Legal exposure arrives fast — incorrect performance ratings, discriminatory hiring decisions, or biased redundancy selections carry substantial legal weight. Employees who believe they have been wrongly assessed by an automated system have both the motivation and legal standing to pursue remedies.
Operational disruption follows quickly. When employees who are actually high performers receive low ratings, they do not quietly accept it. They disengage, stop going above and beyond, and begin looking for other roles. The institutional knowledge those employees carried cannot be replicated by a new hire — the cost of losing them would dwarf any efficiency gain the AI had delivered.
Finally, the HR team's credibility takes a direct hit. Employees do not blame the algorithm. They blame HR — because HR chose the tool, deployed it, and sent the ratings. Rebuilding that credibility takes far longer than the original error took to create.
"Employees who are wrongly assessed do not just want the error corrected. They want accountability. They want someone to say clearly what went wrong, why it happened, and what specific changes are being made."
— Team peopleHum, peopleHum BlogHow to Continue Using AI After It Gets Something Wrong
The mistake organisations make after an AI failure in HR is that they either abandon the tool entirely, or quietly patch the specific error and continue as before. Both responses are wrong. Abandoning the tool ignores the value AI genuinely delivers when governed well. Returning to the status quo without addressing root cause guarantees a different version of the same failure will occur again.
The right response starts with a root cause review — not just which metric was wrongly weighted, but why it was chosen, who approved it, and what testing was done before deployment. Every metric an AI performance tool uses should be validated by HR professionals who understand what the organisation actually values, and by employees who understand what good performance looks like in their specific roles.
Mandatory human review must be implemented for any AI-generated output that affects an employee's career — before it reaches that employee. And there must be a clear, accessible escalation pathway: who does an employee contact to challenge an AI output? What is the review timeline? Who has authority to override the AI? These questions must have specific answers before a tool is deployed, not after a crisis has already landed.
Rebuilding Trust in AI After It Fails
The hardest part of an AI failure in HR is not fixing the tool. It is fixing the relationship with the people the tool affected. For the logistics company, rebuilding trust started with the CEO personally meeting with each of the forty-three employees who had received wrong ratings — acknowledging what happened, explaining the specific cause, apologising directly, and confirming that corrected ratings had been issued.
HR then took three concrete steps communicated to the entire workforce: a mandatory human review log for all AI-generated performance outputs; an employee advisory group formed to review and reshape the AI's metric framework; and an annual AI governance review with results shared publicly across the organisation.
The cost of building governance before deployment is always lower than the cost of rebuilding trust after a failure. The window to learn this without a crisis is open. The question is whether organisations will use it.
AI doesn't cause HR crises — organisations do. When AI produces wrong outputs, it is because metrics were poorly chosen, governance was absent, and nobody reviewed the outputs before they reached employees. That is an organisational failure, not a technology failure.
AI errors are systematic, not isolated. AI optimises precisely for what it was told to optimise for. When those parameters are wrong or incomplete, flawed outputs are delivered consistently and at full scale — which is what makes them so damaging.
The consequences compound quickly. Wrong AI ratings create legal exposure, cause high performers to disengage or leave, and damage HR's credibility — because employees blame HR, not the algorithm. Rebuilding trust takes far longer than the error took to create.
Employees want accountability, not just correction. When wrongly assessed, employees need the organisation to clearly acknowledge what went wrong, explain why it happened, and commit to specific, verifiable changes. Generic apologies deepen the damage.
Three governance requirements must exist before any AI HR tool goes live. Validated metrics reviewed by HR and employees who understand their roles; mandatory human review of every AI-generated output; and a clear, accessible escalation pathway for employees who want to challenge a decision.