Amazon's 15% Manager Ratio and Six Delayering Plans, Checked Against Filings

In September 2024 Amazon asked every senior leadership organisation to raise its ratio of individual contributors to managers by at least 15% by the end of the first quarter of 2025. Over 2025 its total headcount still rose 1.3%, to about 1,576,000. The Amazon manager ratio mandate became the template for a wave of delayering announcements. Very few of them can be checked against a number anyone outside the company can see.

What Amazon's manager ratio mandate actually said

The instruction sat inside a longer message from Andy Jassy, best remembered for ordering staff back to the office five days a week from 2 January 2025. The structural half got less attention, and it is the half that spread.

The operative sentence is short. Amazon was "asking each s-team organization to increase the ratio of individual contributors to managers by at least 15% by the end of Q1 2025". An S-team organisation is one of the large divisions run by a member of Amazon's senior leadership team.

Jassy gave the reasoning in plain terms. As Amazon grew quickly, he wrote, "we have understandably added a lot of managers" and "added more layers than we had before". His stated aim was that "having fewer managers will remove layers and flatten organizations more than they are today". The same message created a "Bureaucracy Mailbox" for staff to report unnecessary process.

Notice what the memo does not contain. It names no number of roles to be removed, it does not mention AI, and it does not promise to report back on whether the target was met. That third omission matters most. A ratio target with no published baseline cannot be audited from outside, and as of this writing Amazon has not disclosed its starting ratio or its finishing one.

That makes the mandate an unusual object for analysis. It was specific enough to be copied, and opaque enough that nobody copying it could see whether it worked.

The span of control arithmetic behind a 15% ratio

Span of control is the number of people who report directly to one manager. Amazon's ratio of individual contributors to managers is a close cousin of that measure, taken across a whole organisation rather than one team. Raising it by 15% sounds like a modest structural tweak. The arithmetic says otherwise.

Take an organisation with 1,000 individual contributors and 125 managers, a ratio of 8 to 1. A 15% increase moves the ratio to 9.2 to 1. There are only two levers available, and the size of each move is fixed by the maths rather than by policy.

Our calculation. The starting ratio is illustrative because Amazon has not disclosed one. The first row holds for any starting ratio: managers must fall to 1 divided by 1.15, or 87.0% of the original count.

The conversion route is the one the evidence points to

The third row is worth dwelling on, because converting a manager moves both sides of the ratio at once. One person leaves the denominator and joins the numerator. That is why it takes only 8 conversions per 125 managers to clear the target, against 16 outright removals.

I think this is the most likely reading of what happened at Amazon and elsewhere. It is the route with no severance cost, no lost institutional knowledge and no public announcement. It is also the route that shows up as nothing at all in a headcount filing.

Why the ratio says little about cost

A ratio can improve while the payroll grows. If Amazon added individual contributors faster than managers through 2025, the ratio would rise with no manager losing a job. Your own board should treat a ratio target as a statement about shape, not about spend, and ask for the payroll line separately.

The delayering ledger: what was announced and what can be measured

Amazon was not first and it was not alone. The table below lists the delayering announcements from large employers since 2023 that we could trace to a company statement, a filing or a named executive. The right-hand column is the one that matters: whether anything measurable followed.

Sources as listed in the references and linked in the text. Citigroup's plan was reported by Bloomberg in October 2023. Microsoft's phrase is the CFO's, from its April 2025 earnings call, as reported by the Associated Press. Meta's is from its Q4 2022 earnings call.

Read down the last column and a pattern appears. Six large employers announced delayering. One, Bayer, published a before-and-after manager count. Google published percentages without a base. The rest published intent.

That is not proof the programmes failed. It is proof that the public record cannot tell you whether they succeeded, which is a different and more useful finding for anyone deciding whether to copy them.

What headcount filings show since the Amazon manager ratio mandate

Annual reports do not break out managers. They do report total employees, and the total is the only number every company in this story is required to publish on a consistent basis. Here is what it shows for the three US tech companies that announced flattening.

Amazon grew, then cut corporate roles

Amazon reported about 1,556,000 employees at the end of 2024 and about 1,576,000 at the end of 2025, a net rise of about 20,000. Most of that workforce is in fulfilment and logistics, so the total says little about corporate managers specifically. That is the honest limit of the filing.

The corporate cuts came after the ratio deadline. Amazon announced about 14,000 corporate reductions in October 2025 and about 16,000 more on 28 January 2026, which the Associated Press called the second round of mass layoffs in three months. Senior vice president Beth Galetti framed both as "reducing layers, increasing ownership, and removing bureaucracy."

The 2 rounds total about 30,000 roles. Against the 1,576,000 total that is about 1.9% of the company. Against the corporate workforce alone it is far larger, but Amazon's 10-K does not report that denominator, so we do not compute a share of it here.

Alphabet and Microsoft did not shrink at all

Alphabet reported 183,323 employees at the end of 2024 and 190,820 at the end of 2025, a rise of 7,497 or 4.1%. That is the same year in which Google told staff it had 35% fewer managers of small teams.

Microsoft's 10-K filings put full-time headcount at about 228,000 at both 30 June 2024 and 30 June 2025. The US share dipped by about 1,000 and the international share rose by about 1,000. The flattening announced in spring 2025 left the total unchanged.

Fewer managers usually means retitled managers, not fewer people

The gap between the announcements and the headcount lines has a simple explanation, and two companies supplied the evidence for it themselves. Removing a manager role and removing a person are different events. Most delayering does the first far more often than the second.

Bayer's numbers show the split

Bayer is the only company in the ledger to publish both sides. On its Q3 2025 earnings call, chief executive Bill Anderson said the group had gone from "about 15,000 managers to roughly 5,000". Over the same period it had let go of about 13,500 employees, including around 5,000 managers.

Put those numbers side by side. The manager count fell by about 10,000. Only about 5,000 managers left the company. That implies roughly 5,000 former manager roles were converted, merged or otherwise taken off the management count without the person leaving. Treat that as an approximation, because both inputs are rounded figures from a call.

Bayer is also the clearest case of what delayering costs. Its January 2024 agreement with German employee representatives, reported by pharmaphorum, covered a group of 101,000 people and promised a structure that would "reduce management hierarchies, eliminate bureaucracy, and speed up decision-making". Anderson told analysts in November 2025 that headcount would keep falling, but not at the rate of the previous 18 months.

Google said the same thing out loud

Google's disclosure carries its own qualifier. Brian Welle, vice president of people analytics and performance, told an August 2025 all-hands meeting that "right now, we have 35% fewer managers, with fewer direct reports" than a year earlier. NBC News, syndicating CNBC, reported the remark. A person familiar with the matter said the figure covered managers of fewer than three people, and that many of them stayed on as individual contributors.

That is the conversion route from the ratio table, confirmed by the company that executed it. In December 2024 Google had already told staff it cut managers, directors and vice presidents by 10%, and a spokesperson said some of those roles were converted rather than eliminated. Neither disclosure gave a base number, so neither can be turned into a count of people.

My view is that conversion is the right default and that it should be described as such. Turning a manager of two into a senior engineer is a sensible correction of title inflation. Calling it a cost programme invites investors and employees to look for savings that are not there.

Where AI enters the delayering story, and where it does not

The popular version of this story says AI lets companies run with fewer managers. Software, the argument goes, now does the coordination and reporting that managers used to do. The idea had an influential early statement. In October 2024 Gartner predicted that through 2026, 20% of organisations would use AI to flatten their structure and eliminate more than half of current middle management positions, as CIO Dive reported. That is a forecast, and no follow-up tally has been published.

Check the company statements against it and AI is mostly absent from the original decisions. Jassy's September 2024 memo is about layers and bureaucracy. Bayer's model is about decision rights. Google's disclosure was about efficiency and leadership share. Citigroup's 2023 plan predates the current AI cycle entirely.

AI arrives later, and at one company. In a 2025 message, Jassy wrote that "in the next few years, we expect that this will reduce our total corporate workforce as we get efficiency gains from using AI extensively across the company". When the October 2025 cuts came, Jassy described them as "culture" and said they were not driven by finances or AI, according to the Associated Press. Even inside one company, the explanation shifted.

I read this as a sequence, not a cause. Flattening began as a correction to pandemic-era hiring and title inflation. AI became the forward-looking justification once it was available. Whether AI makes a wider span of control workable is a separate question, and it is covered in the analysis of what the middle management layer becomes after agents. On the evidence of company statements, AI enabled the explanation more than the decision.

This has a practical consequence for how you read layoff announcements. When a company attributes cuts to AI, check whether it said so at the time of the original decision. The broader pattern of statements that credit AI after the fact is tracked in the review of how companies word AI layoff statements. The claims that do not survive scrutiny are logged in the check of Cloudflare's AI layoff claim against its evidence.

Where this analysis is weakest

This post argues that delayering is mostly a reshaping exercise and that AI was a late addition to the explanation. Here is where that argument is most exposed.

No company publishes its manager count

The central measure, the number of people in manager roles, is not in any 10-K reviewed here. Total headcount is a blunt proxy. Amazon's total is dominated by warehouse and delivery staff, so a large fall in corporate managers could be hidden inside a rising total. The flat-or-rising lines in the headcount chart are consistent with conversion, and they are also consistent with real manager cuts offset by hiring elsewhere. The filings cannot separate the two.

The brief for this post asked for LinkedIn workforce data on manager headcount. We did not use it. Profile-based datasets depend on self-reported titles, lag real changes, and are not published in a form we could open and cite for each company. A post that relied on them would trade one unverifiable number for another.

The counter-case: spans really are widening

There is independent evidence that managers now carry more people. Gallup reported in January 2026 that the average number of people reporting to a manager rose from 10.9 in 2024 to 12.1 in 2025, an increase of about 11% in a year. Gallup linked the shift to organisations "downsizing or consolidating middle management roles while expanding the number of people who report to each remaining manager". The same Gallup study found the median held at about five to six, which means a minority of very large teams drove the average.

So the sceptical reading here could be too sceptical. Something real is happening to span of control across the economy. What the evidence does not show is that it is happening through mass manager layoffs at the companies that announced it loudest, or that AI is what made it possible.

Frequently asked questions

How to test a delayering plan before you copy it

If you run a team of any size, the useful lesson from this ledger is not "have fewer managers". It is that a ratio target is easy to hit and hard to verify, so you need to decide in advance what would count as success.

This week, write down three numbers before any restructure: your current ratio of individual contributors to managers, your total payroll, and the median time from a decision being requested to a decision being made. Then decide which of the three you expect to move. If only the first moves, you have retitled people. That may be the right call, but call it that.

Next, look at the people in manager roles with fewer than 3 reports. Google drew its line there for a reason. Those roles are the cheapest to convert and the least likely to be doing real management, and reviewing them takes an afternoon rather than a reorganisation. For what the wider labour data says about AI and job losses, read the check of AI job loss claims against payroll data before you put a number in front of your board.

Related reading