BMW Trims Management: Speed Comes from Structure

BMW is trimming its management by 20 percent because speed comes from structure first and from technology second: by mid-2027 the BMW Group will reduce the number of its divisions and the management functions assigned to them by 20 percent, with the levels below to follow on a comparable scale. According to a transparent calculation for a mid-sized company with 800 employees, the same cut reduces the number of possible interfaces between units by up to 38 percent, so that fewer questions between divisions travel up to the executive team. The AI agents for vehicle development that BMW announced in the same release on 30 September 2026 only take effect once they work reliably and engineers have enough time to approve their results.

BMW is cutting divisions and management functions by 20 percent by mid-2027

On 30 September 2026, at its Capital Market Day in Munich, BMW announced that it will reduce the number of its divisions and assigned management functions by 20 percent by mid-2027. According to the BMW Group release this will affect the levels below on a comparable scale. The group already agreed a far-reaching adjustment of its personnel structures with the works council in July.

As produktion.de reported on 1 October 2026 citing dpa, the first step concerns the division head level. There CEO Milan Nedeljković justifies the restructuring by saying that the organisation is “simply too big” and that BMW must become smaller and faster. It follows job cuts that swissinfo reported on 29 July 2026 citing dpa. According to that report, around 8,000 jobs worldwide are to go through natural attrition and a voluntary severance programme for employees in Germany outside direct production.

With a 1 to 3 percent margin in its car business, BMW is targeting the cost of management

BMW is streamlining its management in a year in which, by its own forecast, the car business will reach an EBIT margin of 1 to 3 percent. According to the release on the 2025 financial year of 12 March 2026, the margin in the automotive segment was 5.3 percent in 2025, after 6.3 percent in 2024. At that time BMW expected 4 to 6 percent for 2026. On 16 June 2026 the company lowered the corridor to 1 to 3 percent in an ad hoc announcement, mainly because of the weak Chinese car market and high energy prices. In the same announcement BMW said it would take further structural and efficiency measures that will weigh on earnings once in the second half of 2026.

The gap between forecast and target is considerable, because each percentage point of margin corresponds to around 1.2 billion euros of earnings on the segment's 2025 revenue of 117.6 billion euros. From the midpoint of the 2026 corridor to the midpoint of the target there are seven percentage points, which by our own calculation is around 8 billion euros a year at constant revenue. Neither BMW nor the report by ecomento on the margin targets of 1 October 2026 quantifies which part of this gap the streamlining of management is to close.

Graicunas and Jaques justify a management layer by the decisions it makes itself

A management layer is justified if it makes decisions that the level below cannot make, because every additional layer lengthens the path of a decision and creates costs that do not show up in any vehicle. In 1933 the management consultant V. A. Graicunas showed that the number of possible relationships a manager has to keep track of grows faster with each additional direct report than the span itself. With five direct reports he arrives at 100 possible relationships and with six at 222. In “Requisite Organization” (1989) Elliott Jaques derived the number of levels from the time horizon of the decisions made at each level, so that a level is justified if it decides over a noticeably longer period than the level below it.

A cut of 20 percent also raises the average span of control by 25 percent in arithmetic terms, because the same number of people is spread across four fifths of the functions. How many layers a company needs is covered in Spans and layers, how many people a manager can lead effectively in Span of control. How heads of department can prepare to lead more people after delayering is described in Strengthening middle management.

Cutting 20 percent of units reduces interfaces in a mid-sized company by up to 38 percent

A mid-sized company that follows the BMW pattern and reduces its units on two levels by 20 percent each cuts the number of management functions by 20 percent and the number of possible interfaces between units by 36 to 38 percent. The calculation assumes a company with 800 employees without management duties, who work in 40 departments of 20 people each. Every four departments form one of ten divisions. After the cut there are eight divisions and 32 departments of 25 people each, so that 10 of the 50 management functions disappear. An interface is any pair of units on the same level, which with n units gives n times (n minus 1) divided by 2. Between ten divisions there are 45 such pairs, between eight divisions 28. Between 40 departments there are 780, between 32 departments 496.

Part of the coordination moves into the larger units. A department of 20 people has 190 possible coordination paths between any two people, one of 25 people has 300. Across all departments their number rises from 7,600 to 9,600, an increase of 26 percent. For speed this shift is a gain as long as the head of department has time for these questions, because a question within a department ends with its head, while a question between two divisions can travel up to the executive team. How the number of coordination paths grows with every additional person is shown in Team size.

The 38 percent is an upper limit, because in practice few divisions work with every other division. If each unit has a fixed number of partners, say three, the number of interfaces falls in the same proportion as the number of units, which is 20 percent. The effect therefore lies between 20 and 38 percent. It grows when the units that are merged are those that coordinate frequently, because their interface then moves into a single unit with one head. For the selection a map of interfaces is therefore more telling than headcount. How to design interfaces between functions is described in Breaking down silos.

Trimming management saves money when the functions go, while AI takes effect later

Trimming management is a decision on structure whose savings arrive when the functions are removed, so that they depend on no AI system. BMW puts both topics into the same release and calls agentic AI applications a central lever for speed and leaner structures. In concrete terms the release describes AI use in development, purchasing, production, sales and aftersales. None of the sources gives a figure for savings from AI.

The two levers belong in separate plans, because they take effect at different times. Whether decisions are made faster after the cut depends on how the decision rights of the removed functions are allocated, because otherwise they move upwards. An approach for this is described in Clarifying decision rights. The benefit of AI agents only shows once they work reliably in the workflows.

For AI agents in development, BMW engineers are accountable for every approval

At BMW the engineers are accountable for the results of the AI agents, because according to the company these are “always accompanied, reviewed and given final approval” by them. For crash simulations BMW says it combines its engineering data with the training capabilities of Mistral AI.

In her 1983 paper “Ironies of Automation” the psychologist Lisanne Bainbridge described how automation takes routine work away from people and leaves them the difficult cases, for which they lack practice because the routine is gone. An approval is therefore only as good as the approver's ability to judge a result independently. Companies that turn engineers into reviewers of agents must give them time, review criteria and insight into the basis of the results, because otherwise the signature becomes a formality while accountability remains with the company. How much human control an AI system needs in operation is described in Human in the loop.

When fewer managers lead more people while approving more results from agents, their load rises, so it needs to be checked in advance whether the approvers can do this work in the new structure.

Executive teams in mid-sized companies should count interfaces before trimming management

Executive teams in mid-sized companies can derive a six-step approach from the BMW case that starts with a map of interfaces and measures structure and AI against separate metrics:

1. Count interfaces (weeks 1 to 4, head of finance or COO): For every pair of units, record how often it coordinates and how many questions go to the executive team. The measure of success is a map of interfaces with escalations per month.
2. Judge layers by their decisions (weeks 3 to 8, executive team with the head of HR): For every management function, list the decisions it makes itself. The measure of success is the share of functions with their own decision rights.
3. Merge units by interfaces (by month 4, executive team): Units that coordinate frequently are merged. The measures of success are the number of interfaces and the share of coordination that takes place within one unit.
4. Assign decision rights before the cut-off date (by month 5, division heads): The decisions of the functions being removed go to named roles, so that they do not end up with the executive team. The measure of success is escalations to the executive team per month.
5. Regulate AI approvals (from month 4, head of the business unit concerned): For every AI use, define who approves a result, against which criteria and with how much time for review. The measures of success are review time and rejection rate per result.
6. Check the effects separately (month 12, controlling): Report the cost of management, the duration of decisions and the lead time of AI-supported workflows separately. The measure of success is that every improvement can be attributed to one lever.

For companies in Germany co-determination is added. A reduction of management functions on a larger scale can be an operational change, for which management negotiates a reconciliation of interests with the works council and agrees a social plan. For executive staff the executives' committee is responsible where one has been formed. The process is described in Operational change in reorganisations. This overview does not constitute legal advice.

Niehues Consulting combines organisation design and AI management

Niehues Consulting 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 delayering it asks three questions. In Understand: who decides, who delivers, who uses and where are the interfaces? In Build: which units, decision paths and approvals do operations require? In Introduce: are managers and approvers enabled and workflows adapted? 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 many layers and interfaces your organisation needs and where AI use requires a clear approval: book a consultation.