# American Express Job Cuts: Structure Before Automation Author: Marin Niehues, Niehues Consulting · Published: 2026-10-24 · Organisation Design Source: https://www.niehuesconsulting.de/en/blog/american-express-spans-of-control-ai/ American Express is removing management layers and widening spans of control before it automates its processes with AI. This order is the right one for companies whose lead times are dominated by approvals, because automation only shortens the processing of a case, while in organisations with many layers most of the time is spent waiting for approvals. By a calculation with disclosed assumptions, a case with four approval steps becomes 11 percent faster through automation alone and 53 percent faster once two of those steps have been removed. On 23 October 2026 American Express announced that it will cut around 5 percent of its positions worldwide. It is booking a 336 million US dollar restructuring charge. **Key facts at a glance** - On 23 October 2026 American Express reported a pre-tax restructuring charge of 336 million US dollars together with its third-quarter results. - On the call the same day CEO Stephen Squeri said he expects around 5 percent of positions worldwide to be eliminated. The release and the transcript give no absolute number; according to Bloomberg, as cited by HRD America, CEO Squeri put the figure at about 3,500 positions. - CFO Christophe Le Caillec told CNBC on the same day that the aim is to reduce layers and increase spans of control so that American Express can automate. - The cuts come in a quarter with revenue growth of 9 percent to 20.1 billion US dollars and net income up 28 percent. **American Express is fixing its structure first and automating afterwards, because software reproduces a process as it is built, with every approval step and every handover between departments. Companies that first remove approvals and push decision rights down multiply the effect of the automation that follows.** ## American Express is cutting around 5 percent of positions to remove layers before AI On 23 October 2026 American Express announced a restructuring together with its third-quarter results that affects around 5 percent of its positions worldwide. According to the [earnings release filed with the US Securities and Exchange Commission](https://www.sec.gov/Archives/edgar/data/0000004962/000000496226000405/q326exhibit991.htm), the company is booking 336 million US dollars before tax as a restructuring charge for efficiency initiatives. It is funded from part of the 982 million US dollar gain on the sale of its stake in Global Business Travel Group. CEO Stephen Squeri said on the analyst call, according to the [transcript published by The Motley Fool](https://www.fool.com/earnings/call-transcripts/2026/10/23/amex-axp-q3-2026-earnings-call-transcript/), that earlier this year he had asked senior leadership to streamline their organisations, align their teams on the most critical deliverables and reduce layers to speed decision-making. The restructuring is expected to eliminate approximately 5 percent of positions globally. The release and the transcript give no absolute number; according to Bloomberg, as cited by [HRD America](https://www.hcamag.com/us/specialization/change-management/amex-slashes-5-as-it-strips-out-management-layers-to-clear-the-way-for-ai/594799) on 23 October 2026, CEO Squeri put the figure at about 3,500 positions. The [2025 annual report](https://www.sec.gov/Archives/edgar/data/4962/000000496226000080/axp-20251231.htm) shows approximately 76,800 employees at year end, so 5 percent comes to 3,840 by our own calculation. The link to AI came from CFO Christophe Le Caillec. According to [IBTimes](https://www.ibtimes.com/american-express-becomes-latest-company-announce-massive-layoffs-over-ai-boom-3808922) on 23 October 2026, he told CNBC that American Express wants to "reduce layers and increase spans of controls, so that we can automate" and take advantage of AI in the coming quarters and years. HRD America quotes him as saying that you don't want to automate processes that are not optimally structured. **The restructuring in numbers** - around 5 % | of positions worldwide are to go, according to CEO Stephen Squeri - USD 336m | pre-tax restructuring charge in the third quarter of 2026 - 76,800 | employees at American Express at the end of 2025, 5 percent of which is 3,840 by our own calculation - USD 3.7bn | net income in the third quarter of 2026, up 28 percent year on year ## American Express is cutting jobs on 28 percent higher profit to prepare for automation American Express is cutting jobs in a strong quarter because the restructuring is meant to prepare the organisation for the automation of the coming years. Revenue rose 9 percent to 20.1 billion US dollars in the third quarter of 2026 and net income rose 28 percent to 3.7 billion US dollars. With this reasoning American Express reverses a widespread expectation. In October 2024 Gartner predicted that through 2026 one in five organisations would use AI to flatten their structures and eliminate more than half of their middle management positions, [SHRM](https://www.shrm.org/topics-tools/flagships/ai-hi/gartner-ai-predictions-through-2029) reported on 27 November 2024. In that prediction AI is the cause of the flatter structure. At American Express the flatter structure is the precondition for AI. On 30 September BMW announced 20 percent fewer divisions and management functions plus AI agents for vehicle development in the same release, as described in [BMW Trims Management: Speed Comes from Structure](/en/blog/bmw-management-layers-ai-lessons/). At BMW the two levers stand side by side, at American Express one follows the other. ## Automation fails in deep hierarchies because approvals determine the lead time Automation fails in deep hierarchies because it speeds up the work between approvals while the approvals themselves, the handovers between departments and the exceptions remain tied to the layers. An AI agent that prepares a quote with special terms in minutes hardly shortens the lead time if the quote then waits three days for sign-off by the team lead, the head of department and the division head. At the interfaces between departments the system and data format often change along with responsibility. An automation project has to reproduce each of these handovers and thereby fixes them in software, so that the structure is harder to change afterwards than before. Michael Hammer described this relationship in the Harvard Business Review in 1990 under the title ["Reengineering Work: Don't Automate, Obliterate"](https://hbr.org/1990/07/reengineering-work-dont-automate-obliterate), whose thesis is that companies should remove unnecessary steps before they speed up processes with technology. Automated processes handle standard cases and pass special cases to people. If it is unclear who decides on a special case, it moves up the layers to managers who have not planned time for it. How rules and approvals can be thinned out is described in [Reducing bureaucracy in companies](/en/blog/reducing-bureaucracy-in-companies/). A calculation for a quote with special terms shows how far approvals limit the effect. Processing takes six working hours, that is 0.75 working days. The case passes through four approval steps and waits an average of one working day at each. An AI agent takes over 70 percent of the processing time. The simplification removes two of the four steps because their decision right moves to the direct manager. ::: diagramm Automation alone shortens the case by 11 percent, after removing two approval steps the gain is 53 percent - Starting point with four approval steps | 4.75 | 4.8 days - Only processing automated | 4.225 | 4.2 days - Only two approval steps removed | 2.75 | 2.8 days - !Two steps removed first, then automated | 2.225 | 2.2 days Source: Niehues Consulting calculation; lead time in working days with 6 hours of processing, 1 working day of waiting per approval step and 70 percent of processing time automated As long as approvals stay with people, automation can shorten the lead time by at most the share that processing accounts for. In the starting point that is 16 percent, because 4 of the 4.75 days are waiting time. With half a day of waiting per step the limit is 27 percent, with two days it is 9 percent. [Process mining](/en/blog/process-mining/) shows from system timestamps where the waiting time arises. ## With 1,000 employees, a span of 10 instead of 6 saves one layer and 67 managers With a span of control of six, a company with 1,000 employees needs around 167 managers across five layers, with a span of ten it needs 100 managers across four layers. It assumes that the 1,000 employees include managers and the head of the company, that every other person has exactly one manager and that all managers have the same span. The number of managers is then 999 divided by the span. The number of layers is the smallest number of levels across which 1,000 people can be arranged at that span. ::: diagramm If the span of control rises from 6 to 10, a company with 1,000 employees needs 67 fewer managers and one layer less - Span 4, six layers | 250 | 250 - !Span 6, five layers | 167 | 167 - Span 8, five layers | 125 | 125 - !Span 10, four layers | 100 | 100 - Span 12, four layers | 83 | 83 Source: Niehues Consulting calculation; number of managers with 1,000 employees, the same span for every manager and one manager per person Moving from six to ten removes 67 management roles, 6.7 percent of the workforce and the same order of magnitude as the 5 percent at American Express, although it is unknown what share of the cuts there concerns management roles. Every layer removed also takes one approval step out of every case that has to go up to the executive team, so that six layers mean five handovers and four layers mean three. How an organisation determines the right number of layers is covered in [Spans and layers](/en/blog/spans-and-layers-management-levels/). A span of six reflects the situation in many companies. According to [Gallup data from January 2026](https://www.gallup.com/workplace/700718/span-control-optimal-team-size-managers.aspx), the median manager in the United States leads six people and two thirds lead fewer than ten. Whether a larger span works depends, according to Gallup, on how much time the manager has for leading. 97 percent of managers do individual contributor work, a median of 40 percent of their time. Those who spend less than 40 percent on it show higher engagement than the average of 37 percent, regardless of team size. A span should therefore only grow as far as the manager hands off individual work. Here AI has a second use once layers are gone, because it reduces the individual work of the remaining managers and frees time for larger teams. What span is effective is described in [Span of control](/en/blog/span-of-control-direct-reports-per-manager/). ## Before automation, every approval that merely passes a case on should go Before automation every approval that merely passes a case on should go, while rule-based checks are written down and judgement calls are placed as low as expertise allows. An approval step can be tested by how often it changes or rejects a case. If it hardly ever changes anything across 100 cases, it costs waiting time and adds nothing to the decision. **Four types of step in an approval chain** Step | How to recognise it | Before automation | After automation Pass-through approval | The step rarely changes anything in the case | remove | gone Rule-based check | The decision follows fixed limits such as amount or completeness | write the rule down | software checks, people take samples Handover between departments | The case changes responsibility or system | give one person responsibility for the whole process | automate end to end Judgement call | Weighing, liability or exception | place the decision right as low as expertise allows | AI prepares, a person decides The order of the table is the order of the work, because only once pass-through approvals have gone is it clear which checks and handovers remain. How decision rights can be allocated is described in [Clarifying decision rights](/en/blog/decision-rights-raci-rapid/), where AI creates value in processes in [Process automation with AI](/en/blog/process-automation-with-ai/). ## Executive teams in mid-sized companies should cut approvals first, then automate Executive teams in mid-sized companies can derive a five-step approach from the American Express case that starts by measuring waiting time and puts automation at the end: 1. Map approval chains (weeks 1 to 4, commercial director or COO): measure approval steps, processing time and waiting time from the systems for the ten most frequent cases. The success measure is the share of waiting time in the lead time. 2. Remove pass-through approvals (months 2 and 3, division heads): every step that has hardly ever changed anything in the last 100 cases goes. Its decision right moves to the direct manager. The success measure is the number of approval steps per case. 3. Reorganise layers and spans (by month 6, executive team with the head of HR): every layer must make decisions that the layer below cannot make. A span only grows where the manager spends less than 40 percent of their time on individual work. The success measures are the number of layers and the span per manager. 4. Automate afterwards (from month 6, one owner per process): automate rule-based checks and handovers and let AI prepare judgement calls. The success measures are lead time and rework per case. 5. Review the effect (month 12, controlling): the same measures as in step 1 show what share of the reduction comes from structure and what share from automation. For companies in Germany co-determination comes on top. Removing management layers on a larger scale can be an operational change (Betriebsänderung) for which the executive team negotiates a reconciliation of interests with the works council and agrees a social plan. The works council also has a say on software that can record employees' performance or behaviour. The process is described in [Operational change in reorganisations](/en/blog/operational-change-social-plan-germany/). This assessment does not replace legal advice. ## Niehues Consulting combines organisation design and AI management [Niehues Consulting](/en/) supports executive teams and heads of department in building organisations that deliver reliably. Organisation design creates structures and decision paths for day-to-day business and for transformation. In delivery management, a delivery manager or product owner steers the programme until the change has arrived in operations. Where AI enters the workflows, a business case and a prototype are added, followed by introduction in the business units. The approach follows the Niehues Delivery Model with its phases Understand, Build and Introduce. Before automation it asks three questions. In Understand: who decides, who delivers, who uses? In Build: which layers, decision paths and approvals do operations require? In Introduce: are managers enabled and workflows adapted before software reproduces them? Founder Marin Niehues, a management consultant from the delivery side, builds organisations that deliver. His projects range from venture-backed scale-ups to DAX 40 groups, with assignments in more than 20 countries on six continents. In the free 30-minute initial consultation you assess together how much waiting time sits in your approval chains and which steps can go before automation: [book a consultation](/en/#kontakt). **What executives can take from the American Express case** American Express is cutting around 5 percent of its positions to reduce layers and increase spans of control before it automates. Companies that first measure waiting time, remove pass-through approvals and only widen spans as far as managers hand off individual work create the basis on which automation has a measurable effect. **Sources** - [American Express: third-quarter 2026 earnings release, Exhibit 99.1 to Form 8-K](https://www.sec.gov/Archives/edgar/data/0000004962/000000496226000405/q326exhibit991.htm), 23.10.2026 - [The Motley Fool: Amex (AXP) Q3 2026 Earnings Call Transcript](https://www.fool.com/earnings/call-transcripts/2026/10/23/amex-axp-q3-2026-earnings-call-transcript/), 23.10.2026 - [American Express: Annual Report on Form 10-K for fiscal year 2025](https://www.sec.gov/Archives/edgar/data/4962/000000496226000080/axp-20251231.htm), February 2026 - [IBTimes: American Express Becomes Latest Company to Announce Massive Layoffs](https://www.ibtimes.com/american-express-becomes-latest-company-announce-massive-layoffs-over-ai-boom-3808922), with statements by Christophe Le Caillec to CNBC, 23.10.2026 - [HRD America: Amex slashes 5% as it strips out management layers to clear the way for AI](https://www.hcamag.com/us/specialization/change-management/amex-slashes-5-as-it-strips-out-management-layers-to-clear-the-way-for-ai/594799), 23.10.2026 - [Gallup: Span of Control: What's the Optimal Team Size for Managers?](https://www.gallup.com/workplace/700718/span-control-optimal-team-size-managers.aspx), 13.01.2026 - [SHRM: report on Gartner's predictions for AI and organisations](https://www.shrm.org/topics-tools/flagships/ai-hi/gartner-ai-predictions-through-2029), 27.11.2024 - [Michael Hammer: Reengineering Work: Don't Automate, Obliterate](https://hbr.org/1990/07/reengineering-work-dont-automate-obliterate), Harvard Business Review, July 1990 ::: faq Frequently asked questions ### When did American Express announce the job cuts? American Express announced the cuts on 23 October 2026 together with its third-quarter results. ### How many jobs is American Express cutting? American Express states around 5 percent of positions worldwide. The release and the transcript give no absolute number; according to Bloomberg, as cited by HRD America, CEO Squeri put the figure at about 3,500 positions. Based on the approximately 76,800 employees at the end of 2025, our own calculation gives 3,840. ### Why is American Express removing management layers before it automates? CFO Christophe Le Caillec told CNBC that you don't want to automate processes that are not optimally structured. The company is therefore reducing layers and increasing spans of control first so that it can then automate and take advantage of AI. ### Is American Express replacing the jobs with AI? American Express justifies the cuts with a leaner structure and faster decisions. The company has not published how much work AI is to take over or what savings it expects. ### Why does automation achieve little in organisations with many approvals? Automation only shortens processing time, while in deep structures cases spend most of their lead time waiting for approvals. In a sample calculation with four approval steps and one day of waiting per step, processing accounts for only 16 percent of the lead time. ### What can German companies learn from the American Express case? German companies can learn to measure the waiting time in their approval chains before automating and to remove pass-through approvals. Spans should only grow as far as managers hand off individual work. For larger reductions in layers and for software that can record performance, they involve the works council early.