Stack-Ranking Came Back. It Just Changed Its Name.

A decade ago, the verdict looked settled. GE retired Jack Welch’s vitality curve. Microsoft killed stack ranking in 2013 after Vanity Fair linked it to the company’s lost decade. Adobe, Deloitte, Accenture, Gap; one after another, the big names walked away from forced rankings and announced a softer future of continuous feedback and growth conversations. We told ourselves the bell curve was finished.

It wasn’t. It went quietly, picked up new vocabulary, and came back.

Look at what’s happened since 2022. Goldman Sachs reinstated a three-tier forced distribution and resumed cutting its bottom performers. Amazon reportedly runs something called “Unregretted Attrition,” a roughly 6% annual target that feeds a performance-improvement track expecting about a third of its entrants to leave. Then there’s Meta. In January 2025, the company announced cuts to around 5% of its workforce, and internal guidance reportedly told managers on teams of 150 or more to label 15 to 20% of their reports “below expectations,” up from the prior 12 to 15%. When real underperformers were scarce, managers were allowed to reclassify people who’d been rated at or above expectations just to hit the number. Some of the employees shown the door had strong reviews on file.

None of these companies calls this stack ranking. The language is all “raising the bar,” “low performers,” “good attrition.” But strip the branding, and the mechanics are identical to what we supposedly buried in 2013: a pre-set percentage of your people will be graded against each other and sorted into a losing tier, whether or not their actual work warrants it.

The simple fact is that a quota with a friendlier name is still a quota.

So we should ask the obvious question. Did the evidence change? Did someone discover that ranking people against each other actually works better than we thought?

No. The research base is right where it was, and it points the other way.

The foundational study here is Kluger and DeNisi’s 1996 meta-analysis—more than 600 effect sizes across some 23,000 observations. Feedback helped performance on average, but more than a third of the interventions they studied actually made performance worse. What separated the feedback that helped from the feedback that hurt? Attention. When feedback pointed people toward the task—what to do differently, how to improve the work—it tended to help. When it pointed people toward themselves, “Where do I rank? Am I good enough? “ Am I safe?” It tended to backfire.

A 25-year review of feedback research by Anseel and Sherf, published in 2025, doesn’t soften that picture. Their honest summary is that the science of workplace feedback still isn’t a clean story of progress, and that what researchers know stays oddly disconnected from how organizations actually run feedback. That’s not a green light for confident ranking systems. It’s a cautionary tale for anyone who tells you their forced distribution is “data-driven.”

There’s a newer wrinkle the old debate didn’t have: artificial intelligence. Most large employers now track performance digitally, and a majority report using AI tools to evaluate it. The trouble is that these systems generate comparative rankings as a byproduct, a sorted dashboard a manager can scroll, even when leadership has formally sworn off forced distribution. One NYU researcher described Amazon’s automated attrition machinery as the old stack ranking with a new, automated twist. You can abolish the annual bell curve in a town hall and rebuild it by accident inside your productivity software. If your tooling produces a ranked list that your managers can see, you’re running stack ranking.

This is the gap between strategy and execution, showing up in a particularly costly form. Most leaders, asked directly, will say they want to develop their people, build their bench, and keep their best talent. The intent is genuine. But the system they’ve installed—or let creep in through the monitoring stack—teaches a different lesson every single week: protect your standing, watch your back, and don’t help the colleague who might outrank you. Relationship infrastructure either supports development or quietly corrodes it, and a ranking machine corrodes it.

None of this means you stop differentiating performance. Every organization needs to know who’s carrying the load and who isn’t. But three things get tangled together that shouldn’t be: differentiating performance, ranking people against each other, and removing genuine underperformers. You can do the first and the third without the second. Absolute standards, “here’s the bar, here’s whether you cleared it,” accomplish nearly everything proponents claim ranking does, without pitting your team against itself.

A few checks are worth running on your own organization.

Start by auditing the language. Walk through how your company decides on terminations and reductions. If anywhere in that process there’s a pre-set percentage—”bottom 5%,” “15% below expectations,” an attrition target—you’re running forced distribution no matter what the deck calls it. Name it honestly to yourself first.

Next, run the attention test on your feedback. Pull a handful of recent performance conversations and ask one question: Did they direct people toward the work or toward their rank? If your managers are mostly telling people where they place rather than how to improve, you’re in a regime where the evidence says feedback backfires a third of the time.

Then check your monitoring tools. Ask your people team a direct question: does any dashboard a manager can open produce a ranked ordering of employees? If so, you’ve reintroduced the bell curve through the back door, and you need to decide, for a purpose, whether that’s what you want.

Finally, watch your top-quartile attrition. If your best people are leaving at a rising rate, the ranking system is a prime suspect—and rolling it back tends to be the highest-return move available. When Adobe killed its stack-ranked review, voluntary turnover dropped sharply within a year.

The elite signal in tech and finance is swinging back toward the bell curve. Before you follow the crowd, make sure you know which one you’re trusting.

This work synthesizes research from Kluger and DeNisi’s foundational meta-analysis on feedback interventions (Psychological Bulletin, 1996), Anseel and Sherf’s 25-year review of workplace feedback research (Annual Review of Organizational Psychology and Organizational Behavior, 2025), and Van Strydonck and colleagues’ study on comparative feedback and counterproductive work behaviour (European Journal of Work and Organizational Psychology, 2026), alongside documented performance-management practices at Meta, Amazon, Goldman Sachs, Microsoft, and Adobe.

Contact us for a complete list of works cited.