Microsoft's Lost Decade: How Stack Ranking Killed Collaboration

8 minute read

Kurt Eichenwald posed a thought experiment in his 2012 Vanity Fair investigation of Microsoft: imagine the company had hired Steve Jobs, Mark Zuckerberg, Larry Page, Larry Ellison, and Jeff Bezos onto a single team before they made their names elsewhere.

Under Microsoft's stack ranking system, two of them would have to be rated as below average. One would be deemed disastrous.

That's not a performance system. That's manufacturing failure.

For a decade, Microsoft ran a performance management process that every employee Eichenwald interviewed called "the most destructive process inside of Microsoft." The system had a simple premise: every team would identify their top performers, their average performers, and their bottom 10%. Even if everyone was excellent. Even if the team was full of future billionaires.

The math guaranteed it.

The Lost Decade

Between 2000 and 2010, Microsoft stagnated while competitors surged. Despite strong revenues, shareholder returns were largely flat, and the company significantly underperformed its peers.

Over the same period, Apple’s stock rose roughly twenty-fold, and in later years iPhone revenue alone would surpass what Microsoft earned annually during that era. Microsoft repeatedly found itself on the wrong side of major platform shifts: its early e-reader efforts lost to Amazon’s Kindle; Zune was eclipsed by the iPod; Windows Mobile fell behind iOS and Android; billions spent on search still couldn’t displace Google; and the company never built a meaningful social platform as Facebook exploded.

This wasn't a technology failure. Microsoft had resources, talent, and market position. The company was hiring brilliant engineers. The R&D budget was massive.

What failed was organizational. And at the center of that failure was a performance system that sounded perfectly logical: identify your best people, reward them heavily. Identify your worst, help them improve or move them out. Force managers to differentiate rather than giving everyone high ratings.

Stack ranking. The system Jack Welch had used to transform GE in the 1980s and 1990s. The system that dozens of Fortune 500 companies adopted because if it worked for the most celebrated CEO of his era, it should work everywhere.

The logic was sound. The implementation was predictable. The results were catastrophic.

How Stack Ranking Worked

Every unit at Microsoft was forced to declare a certain percentage of employees as top performers, good performers, average, below average, and poor. The specific percentages varied over the years, but the principle remained: someone had to be at the bottom.

If your team had ten people, you walked in the first day knowing that no matter how good everyone was, the distribution was predetermined. Two people would get great reviews. Seven would get mediocre reviews. One would get a terrible review.

Top ratings meant bonuses, stock options, promotions. Bottom ratings meant no cash bonus and usually an exit. Middle ratings meant you were told "3 really isn't that bad" while watching your career stall.

Your rating didn't reflect your actual performance. It reflected your performance compared to your immediate teammates.

The system was relative, not absolute. Your rating didn't reflect your actual performance. It reflected your performance compared to your immediate teammates.

I see this logic when I work with scaling companies. Leaders want to avoid grade inflation. They don't want every manager giving everyone high ratings. So they force differentiation through curves or rankings. It feels rigorous. It feels fair.

Then the behavioral responses start.

What Employees Actually Did

Eichenwald's investigation, based on dozens of interviews with current and former Microsoft executives and thousands of internal documents, found a universal pattern.

Employees avoided working with top performers. The logic was inescapable: if you're rated relative to your teammates, working with excellent people makes you look worse by comparison. Microsoft superstars did everything they could to avoid working alongside other top-notch developers, out of fear they would be hurt in the rankings.

Rational response: seek out weaker teams where you'll rank higher. Result: talent dispersed rather than concentrated on critical projects. The all-star teams that might have built breakthrough products never formed because forming them would guarantee some stars got marked as failures.

Information hoarding became strategy. One employee described the pattern directly: "People responsible for features will openly sabotage other people's efforts. One of the most valuable things I learned was to give the appearance of being courteous while withholding just enough information from colleagues to ensure they didn't get ahead of me on the rankings."

Helping colleagues improve their work hurt your own ranking. Withholding information created competitive advantage. The appearance of collaboration without substance became the optimal strategy.

Short-term focus replaced innovation. Because reviews came every six months, employees and their supervisors (who were also ranked) focused on their short-term performance rather than longer efforts to innovate. The best way to guarantee a higher ranking was to impress not only your boss but bosses from other teams as well. Politics replaced product focus.

Microsoft knew this was happening. Employee surveys conducted every six months delivered the same message repeatedly: groups at Microsoft that are supposed to be working together aren't, a symptom of the stack ranking program. The company did nothing.

Ask yourself: if helping your colleague succeed could hurt your career, would you help them? If hoarding knowledge gave you competitive advantage in your performance review, would you share it? These aren't moral failures. These are rational responses to the incentive structure.

Why This Response Is Predictable

In 1981, economists Edward Lazear and Sherwin Rosen published a paper on rank-order tournaments as labor contracts. Their insight: when compensation is based on ordinal rank rather than absolute output, workers compete against each other instead of cooperating.

Tournament structures can be efficient when measuring output is difficult but ranking is easy. Prize spread (the difference between winner and loser rewards) determines effort level. Make the top prize big enough, and people work harder to win it.

The theory had a critical assumption: workers compete through individual effort. It didn't account for situations where helping others is part of the job.

Innovation and knowledge work require collaboration. Features need cross-team coordination. Complex problems need diverse expertise. Breakthrough products emerge from people building on each other's ideas.

Stack ranking turned all of that into a liability. Every colleague became a competitor. Every conversation became a calculation. Every piece of shared knowledge was a potential loss in the zero-sum game the performance system had created.

The psychology behind this had been understood since 1949, when Morton Deutsch published his theory of cooperation and competition. He showed that goal interdependence determines behavior. When goals are structured so that one person's success helps others succeed (cooperative), people share resources and information. When goals are structured so that one person's success makes others fail (competitive), people hoard and sabotage.

Stack ranking created pure competitive interdependence. Your success required your teammate's failure. The system selected for behaviors that optimized individual ranking at the expense of collective outcomes.

Leon Festinger's research on social comparison explains what happens next. When evaluation is relative, everyone becomes a threat. Upward comparison (to better performers) triggers anxiety and self-protection. Rather than learning from top performers, you avoid them. Rather than seeking out excellence, you seek out mediocrity that makes you look good by contrast.

Microsoft's stack ranking wasn't creating these dynamics through bad implementation. These dynamics were mathematically guaranteed by the structure itself.

The Product Consequences

Innovation requires collaboration across boundaries. Different teams need to work together. Engineers need to share knowledge. Product managers need to coordinate roadmaps.

Stack ranking punished exactly this behavior.

Features that required cross-team coordination stalled because nobody wanted to risk their ranking by helping another team succeed. The best talent avoided the hardest problems because tackling them meant working with other top performers who would make them look bad. Information silos formed because sharing knowledge meant giving away competitive advantage.

The irony: the system meant to drive performance destroyed the conditions performance requires.

Bill Gates saw it happening. A 2003 email revealed his frustration: Microsoft was going to be "so late with a music service" that they'd be "behind others almost forever." The company wasn't lacking ideas or resources. It was lacking the organizational conditions that let ideas become products.

The Resolution

Microsoft eliminated stack ranking in November 2013. Lisa Brummel, Microsoft's HR chief, sent an internal memo: "No more curve, no more ratings."

The timing wasn't coincidental. Satya Nadella became CEO in February 2014. He brought Carol Dweck's growth mindset research into Microsoft's culture transformation. The shift wasn't just about performance systems. It was about changing the fundamental question from "How do I rank?" to "How do I improve?"

Microsoft's resurgence followed. Azure became a cloud powerhouse. The company's market cap surpassed $3 trillion by 2024. Innovation accelerated. Employee engagement improved. The company that had stagnated for a decade became one of the most valuable and dynamic companies in the world.

Stack ranking wasn't the only factor in the Lost Decade. Leadership decisions, market timing, and strategic bets all mattered. But the elimination of forced ranking was part of a successful transformation that proved the counterfactual: when you remove systems that pit teammates against each other, collaboration becomes possible again.

What This Means for Your Organization

Stack ranking is rare now. GE abandoned it after Welch. Adobe, Accenture, Amazon, and dozens of other companies eliminated forced curves. The practice has been so thoroughly discredited that few companies would implement it today.

But the principle behind stack ranking (relative evaluation creating zero-sum dynamics) shows up in subtler forms.

Four questions diagnose whether your system accidentally creates the same incentives:

Does helping a colleague improve hurt your evaluation? If your bonus pool is fixed and shared with teammates, helping them perform better means less for you. If your promotion depends on being top-ranked in your cohort, their success is your loss. The structure doesn't have to be called "stack ranking" to create competitive dynamics.

Would hoarding information give you advantage in your review process? If being seen as the expert improves your rating, sharing expertise dilutes your advantage. If your metrics depend on being the person who solved the problem, collaborating on solutions splits the credit. These aren't hypothetical calculations. People make them constantly.

Do people choose projects or teams based on who makes them look good? If working with strong performers makes you look worse by comparison, the rational choice is to avoid them. If your team's median performance becomes the baseline you're judged against, weak teams become attractive. Watch where your top performers cluster. If they're not working together, ask why.

If everyone optimized for their individual metrics, would the organization succeed? This is the test. Map your individual performance metrics. For each one, ask: what behavior does optimizing this metric produce? Now ask: when Person A optimizes their metric, how does it affect Person B's ability to optimize theirs? If individual optimization conflicts with collective success, you have a stack ranking problem even if you don't have stack ranking.

Microsoft's experience shows that you don't need bad people or bad intentions to create organizational dysfunction. You just need a system where rational individual responses produce irrational collective outcomes.

The math does the rest.

Stack ranking was eliminated because the evidence became impossible to ignore. But the lesson isn't "don't do stack ranking." The lesson is: performance systems create behavior. If you design a system where success requires your teammate's failure, you'll get sabotage and hoarding. If you design a system where success requires your teammate's success, you'll get collaboration.

The choice isn't between rigor and leniency. It's between systems that align individual and collective success, and systems that pit them against each other.

Microsoft spent a decade learning which one works.

---

References

Deutsch, M. (1949). A theory of cooperation and competition. Human Relations, 2(2), 129–152. https://doi.org/10.1177/001872674900200204

Dweck, C. S. (2006). Mindset: The new psychology of success. Random House.

Eichenwald, K. (2012, August). Microsoft’s lost decade. Vanity Fair. https://www.vanityfair.com/news/business/2012/08/microsoft-lost-mojo-steve-ballmer

Festinger, L. (1954). A theory of social comparison processes. Human Relations, 7(2), 117–140. https://doi.org/10.1177/001872675400700202

Lazear, E. P., & Rosen, S. (1981). Rank-order tournaments as optimum labor contracts. Journal of Political Economy, 89(5), 841–864. https://doi.org/10.1086/261010

Nadella, S., Shaw, G., & Nichols, J. T. (2017). Hit refresh: The quest to rediscover Microsoft’s soul and imagine a better future for everyone. Harper Business.

Welch, J., & Welch, S. (2005). Winning. Harper Business.

Wingfield, N. (2013, November 12). Microsoft abandons its employee ranking system. The Seattle Times. https://www.seattletimes.com/business/microsoft-ditches-system-that-ranks-employees-against-each-other/