In brief:
- GE popularized forced ranking in the 1980s, then ditched it in 2015, calling it more ritual than results.
- Meta’s “lowest performers” framing in early 2025 became “AI restructuring” within a year, but a 2026 lawsuit claims performance ratings were never dropped from the selection criteria.
- Companies that introduced a real, visible performance bar (Adobe, Deloitte, GE) saw voluntary attrition fall and feedback quality rise, not the reverse.
- None of this argues that performance should go unmanaged. It argues for naming the bar honestly instead of hiding a quota behind a new label.
Reward the Bar, Not the Curve: The Quota That Keeps Getting Renamed
In the LinkedIn post that goes with this article, I asked whether an employee can actually name a bar they can clear and trust, or whether the real bar is a quota – a fixed percentage cut no matter how the year went. A short post can only point at the evidence behind that question. I expand the topic in this article: where the curve came from, how it keeps resurfacing under a new name even after a company says it’s gone, and what a handful of companies did when they replaced it with something people could actually trust.
Where the curve came from
General Electric made forced ranking famous in the 1980s under Jack Welch: every year, rank the top 20%, call the middle 70% average, and cut the bottom 10%, regardless of how the company as a whole had performed. The logic was that a company gets sharper by continuously shedding its weakest layer. For a long period “Welch’s curve” remained something that other companies copied wholesale.
GE itself walked away from it in 2015. Susan Peters, then the company’s head of HR, told Fortune the old process had “become more a ritual than moving the company upwards and forwards.” It was replaced with an app called PD@GE: continuous short-term goals, on-demand feedback either person could request at any time, and frequent manager 1:1 conversations instead of one dreaded annual sit-down. The year-end meeting didn’t disappear, but its purpose changed from handing out a ranking to coaching on what came next.
Microsoft’s version and its end
Microsoft ran its own forced-curve system for roughly a decade. CNN Money reported that CEO Steve Ballmer scrapped it in November 2013, weeks before Satya Nadella took over. The move was preceded by years of complaints that it pushed employees to compete with each other instead of with the market. An internal memo from HR chief Lisa Brummel confirmed the change at the time.
Microsoft’s 2025 round of cuts was framed by the company as running off an individual performance bar rather than a reinstated curve. That is Microsoft’s own account of its process, not something independently verified through documents the way Meta’s criteria were (see below.) I therefore treat it here the same way I did in the LinkedIn post: as the company’s stated position, not an independently verified fact.
The quota that keeps getting renamed
Amazon’s version of the same idea surfaced in April 2021, when Business Insider published an internal AWS memo describing a target called “unregretted attrition”: roughly 6% of the office workforce pushed out annually, whether through resignation or termination. The Seattle Times corroborated the broader pattern two months later. Employees who landed at the bottom of a ranking entered a program called Focus – a coaching track for underperformers. If that didn’t resolve things, they moved to Pivot, where the choice was a formal improvement plan or a severance package. One manager quoted in the reporting described the feeling as being let go for an A- when everyone else in the room got an A. Amazon’s spokesperson denied that the company stack-ranks employees, while not disputing that the leaked documents described a real internal policy. Fortune reconfirmed the broad shape of the practice again in 2024.
Meta’s version is the clearest illustration of how a quota survives a rebrand. In January and February 2025 Meta cut roughly 5% of its workforce, explicitly targeting what it called its lowest performers. By May 2025, an internal memo raised the share of the workforce expected to land in the “below expectations” bucket from 12-15% to 15-20% for larger teams, ahead of that year’s midyear review. In February 2026, Meta said publicly it wasn’t planning to repeat the scale of those cuts. Then, in May and July 2026, roughly 8,000 more roles were cut under a new label: “AI restructuring.”
That relabeling ended up being tested in court. In July 2026, 26 current and former Meta employees filed a bias lawsuit in the Northern District of California (case No. 3:26-cv-07122), alleging that a new “AI usage” metric was factored into who was cut, and that it penalized people who’d been on medical or family leave and couldn’t rack up the same usage numbers. Meta’s own court filing denied that AI selected anyone on its own, but it confirmed that historical and recent performance ratings were still among the documented criteria used, along with job level, tenure, location, job function, and reporting structure. A federal judge declined to block the layoffs that same month, finding no irreparable harm, but left the door open to revisit the question if more evidence on AI’s role surfaced. Because Meta’s employment agreements require it, the 26 plaintiffs’ individual arbitrations had already started in parallel with that court fight. A hearing on the plaintiffs’ separate bid to block the layoffs outright was held in late August 2026; no public ruling had surfaced as of this writing, and arbitration proceedings aren’t public either.
The allegation about the AI-usage metric penalizing leave-takers is the plaintiffs’ claim, not an established fact. What Meta’s own filing confirms is narrower but it’s enough to settle the question I opened this article with. The framing has moved from “lowest performers” to “AI restructuring.” The performance rating that underpins it hasn’t moved at all.
What a trustworthy bar looks like
I’m not arguing that performance shouldn’t be managed or that nobody should ever be let go. I’m advocating for naming the bar honestly. A few companies have done that and backed it with real numbers.
Adobe dropped its annual ranking system in 2012 in favour of what it called Check-in: frequent, informal conversations between managers and employees with no fixed template and real discretion over remuneration tied to actual goals instead of a forced distribution. Led by then-EVP Donna Morris, the change produced results Adobe was happy to publish: voluntary attrition fell 30%, involuntary departures (the ones Adobe considered overdue) rose 50% because managers stopped saving difficult conversations for that one dreaded day a year, and 78% of employees said their manager was genuinely open to feedback – a real improvement over Adobe’s own earlier surveys. The old system had also been consuming ca. 80,000 hours a year across 2,000 managers, the equivalent of 40 full-time employees doing nothing but paperwork and ranking meetings.
Deloitte’s version of the bar, documented by its own leaders in a widely cited 2015 Harvard Business Review article, started from a similar discovery: its old review cycle was consuming close to 2 million hours a year in forms, meetings, and ratings. Its replacement consisted of three parts: a quarterly snapshot where team leads answered a short set of forward-looking questions about each person rather than scoring them against peers (would I want this person on my team again, are they ready for a promotion, are they at risk of underperforming, and what would I actually pay them if it were my own money); weekly check-ins focused on current work; and a self-assessment tool for employees to name their own strengths rather than wait to be told them once a year.
GE’s PD@GE and Susan Peters’ comment above belong in the same list. A company that had spent three decades as the most famous practitioner of the forced curve walked away from it in the same decade that Adobe and Deloitte did, for very similar reasons.
None of these three is a case of a company getting softer. Each company has ended up reporting fewer surprises, fewer avoidable departures, and less time spent on a ritual that wasn’t improving anything.
Where this leaves the bar
Having a trustworthy bar doesn’t mean nobody is ever let go, and none of the three companies above claims that it does. It means the number of people made redundant isn’t fixed before anyone’s actual work gets reviewed. It also means that a company is willing to say what the bar actually is rather than relabeling the same quota every time the previous name draws scrutiny. That’s the test I’d apply to any company’s next round of cuts, whatever this year’s label turns out to be.
Next in this series: E, Exhaust the Ladder Before the Exit, and the cost levers that are supposed to come before any of this.
Sources
- GE’s 2015 reversal and Susan Peters’ quote: Fortune, August 13, 2015
- Microsoft ending stack ranking under Ballmer: contemporary coverage at GeekWire, November 13, 2013 (CNN Money broke this story at the time; its own page no longer loads)
- Amazon’s “unregretted attrition” memo, Focus and Pivot: Business Insider, via HCAmag, April 24, 2021, corroborated by Human Resources Online, citing the Seattle Times, June 24, 2021 (includes Amazon’s denial)
- Amazon practice reconfirmed: Fortune, March 20, 2024
- Meta’s May 2025 “below expectations” expansion: People Matters, May 22, 2025
- Meta’s February 2026 statement: People Matters, February 16, 2026
- Meta’s May 2026 “AI restructuring” cuts: Reuters via CP24, May 20, 2026
- The July 2026 Meta bias lawsuit and court filing: Bloomberg Government/Daily Labor Report, July 28, 2026; ruling covered in People Matters, July 20, 2026; also covered by cxm.world, July 27, 2026
- The 26 individual arbitrations proceeding alongside the court case, and the August 24, 2026 hearing date: Lumen Law Center, July 17, 2026
- Adobe’s Check-in program and figures: Chartered Management Institute, “Why Adobe killed off the annual performance review”
- Deloitte’s performance management redesign: Buckingham and Goodall, “Reinventing Performance Management,” Harvard Business Review, 2015
