The C-Suite's Blind Spot: What the BMW Restructuring Reveals from an AI Perception Management Perspective

The numbers are well known, but the question is a different one.

BMW is cutting around 8,000 jobs worldwide by the end of 2027, primarily in indirect areas: administration, development, purchasing, and communication. Production is exempt because the group needs every hand on the assembly line for the ramp-up of the Neue Klasse. According to the company, the cuts will be carried out voluntarily through natural fluctuation, partial retirement, and severance packages. The works council and executive board negotiated the program together; compulsory redundancies are not planned. The return on sales in the automotive segment recently stood at 2.3 percent, with a long-term target of 8 to 10 percent. The new leadership team around CEO Milan Nedeljković is intended to prevent BMW from having to write down its profit targets once again.

Much has been written about the causes: China, electromobility, Asian competition, and cost structures. I consider these explanations to be correct. However, I consider them incomplete - specifically in an area that almost never appears in management debates.

My thesis is this: Today, most executive boards can precisely state what the stock market thinks about their company. Hardly anyone can state what the machines say about their company - machines that investors, banks, job applicants, and business partners increasingly use to conduct research. This is not a marginal footnote. In my view, it is one of the biggest open governance gaps of this decade.

Why Classical Early Warning Systems Reach a Limit

According to manager magazin, Chief Financial Officer Stefan Richmann is expected to detect an impending industry upheaval early on the capital market and sound the internal alarm. This mandate is sensible and, in this form, standard: key figures, stock prices, interest rates, analyst consensus, and competitive comparisons. There is nothing wrong with this, and I do not attribute naivety to anyone. Anyone who steps into an office and takes this mandate seriously deserves trust.

The problem lies in the construction of the system, not in the individuals. Classical early warning observes results: figures that others have derived from their assessments. It sees the judgment of the market, not the process in which it originates. And that process has changed.

Anyone wanting to assess a company today increasingly asks an assistant, uses an answer engine, or employs a research tool that synthesizes reports, news, and documents. Technically, this is driven by large language models, often combined with Retrieval-Augmented Generation (RAG) processes that retrieve current sources and incorporate them into the answer. The result is text that sounds authoritative and shapes the first impression. That first impression determines which questions are pursued further and which are dropped.

An early warning system that blanks out this layer only observes the tip of the iceberg. I am not saying that makes it worthless. I am saying it fails to see a growing portion of the information space.

What AI Perception Management Is

AIPM refers to the systematic monitoring, analysis, and curation of how a company is represented in AI-supported information systems. I work with a simple metaphor: Every company has an operational reality - meaning figures, products, strategy, and people. And it has an algorithmic perception - meaning the picture that systems piece together from that reality. Ideally, both coincide. In practice, a gap opens up between them, and this gap is the subject of AIPM.

It arises from several sources:

• Time lag: Models contain knowledge from the past. A company that has transformed can still be described with old characteristics for months.
• Source asymmetry: Critical and alarming reports are shared, cited, and linked more frequently than sober progress reports. In a summary, what is repeated often carries more weight.
• Structural gaps: What a company knows internally but does not publish clearly, currently, and in a machine-readable format does not exist for a system.
• Error inheritance: Incorrect figures, mix-ups, and outdated reports are carried forward if no one corrects them.

I refer to the thematic fields in which these effects occur as Risk Fields: sales forecasts in individual markets, delays in platforms and product launches, management changes, restructurings, legal disputes, and supply chain issues. Each Risk Field has its own narrative on the web, and that narrative can detach itself from the facts.

What This Has to Do with BMW

I want to be precise: I am not claiming that the restructuring at BMW is a consequence of poor algorithmic perception. The causes lie in the market, in costs, and in technology. No amount of polished public image would have changed the situation in China.

What I am claiming is this: BMW serves as a prime case study for how many Risk Fields can be open simultaneously. Analysts had warned against overly optimistic assumptions regarding China. The Neue Klasse is a bearer of hope and a source of uncertainty at the same time, because the market launch and commercial success are still ahead. There was a profit warning a few weeks after the CEO change, alongside a significantly lower stock price, job cuts, and now a restructured leadership team. Each of these events generates texts, commentaries, analyses, and headlines. Combined, they create a narrative that a language model synthesizes.

The decisive question for an executive board is not whether this narrative is unpleasant. It is: Is it accurate, complete, and up to date? For example, are the concentration of job cuts on indirect areas and the protection of production accurately reflected? Is the voluntary nature mentioned, or does the program turn into a blanket "slash-and-burn" narrative? Is the savings goal of around one billion euros annually starting in 2028 contextualized, or is it missing? Such differences shape how outsiders perceive the company, and they can be measured.

Invisible Churn: Silent Attrition

A term I use in this context is Invisible Churn. This refers to the quiet erosion of attention, trust, and - in case of doubt - capital, which is not triggered by a scandal or a wave of protests, but by a multitude of small preliminary decisions: A job applicant rules out an employer after an AI answer describes them as "in transition with an uncertain future." An analyst prioritizes another company because its summary is clearer and more up to date. A procurement manager sees a risk that in reality has long been addressed.

I want to be honest here: How large the share of such effects is in actual capital and personnel decisions cannot be quantified cleanly today. Reliable studies are scarce. My assessment as a practitioner is that the effect is real, is growing with the proliferation of tools, and should therefore be taken seriously before it becomes measurable. Precisely because Invisible Churn is invisible, it does not show up in any classical metric until it is too late. That is an argument for observing it, not for ignoring it.

An Early Warning System That Sees Both Layers

What would I suggest to a Chief Financial Officer whose mandate is to recognize industry upheavals early? I would supplement the classical early warning system with a second layer.

The first layer remains in place: key figures, capital markets, competitors. The second layer regularly answers and documents these questions: How do leading AI systems describe our company? Which topics dominate, and which are missing? Where do the answers deviate from our facts? How does the tone change over time, and how does it compare to competitors?

The crucial factor is change over time. A single test run says little. A series over weeks and months reveals trends: A Risk Field that suddenly appears more frequently and more negatively can be a signal that the public debate is shifting - often before that shift is reflected in numbers. Whoever has this indicator gains time, and time is the scarcest resource in crisis management.

The Perspectives of Those Involved

A topic like this looks different from every role, and I believe it is important to take the objections seriously.

• The Executive Board will say: We have more important things to do. The operational turnaround, the launch of the Neue Klasse, and workforce trust take priority. That is true. My answer: Monitoring is possible with minimal effort, and it protects precisely those priorities because it shows whether progress is reaching the outside world.
• The Works Council and Employees experience the restructuring as a profound cut into their life planning. For them, it is crucial that the job cuts are voluntary and that employment security is extended. An expert article on perception must not downplay this. On the contrary: Especially for employees, it is important that the image of their employer is accurate, because their own future depends on it.
• Investors and Analysts rely on their own models and primary data, and rightly so. Their skepticism is: Where is the proof that general AI summaries move capital? I take this objection seriously. My answer: Professional tools use the same techniques, just with better sources. And the initial orientation - the pre-selection of who is examined more closely - frequently begins with a synthesized answer.
• Communications and PR Professionals face the biggest transformation. I hold a clear position here: Classic PR is not obsolete, but its success metric is changing. Reach and media resonance alone do not indicate whether the resulting machine-generated image is correct. Whoever delivers substance - clear, substantiated, and consistent - has the advantage. Whoever merely serves tone and buzzwords loses impact.
• Journalists provide the raw materials from which summaries are generated. This increases the importance of thorough research and contextualization. I explicitly do not view AIPM as an adversary of journalism, but as a consumer of its results. A company that provides its facts cleanly facilitates accurate reporting.

What Companies Should Do Concretely

From my perspective, there are five steps that any major company can begin without a massive project:

1. Inventory: Ask the questions that an investor, a job applicant, or a business partner would ask of the common systems - in multiple languages and with different formulations. Document the answers and compare them with your facts. Even this simple step brings surprising insights to light for most companies.
2. Classify Errors and Gaps: Not every deviation is equally important. Distinguish between factually incorrect, outdated, incomplete, and one-sidedly weighted. Each category requires a different response.
3. Expand the Fact Base: Publish strategy, key figures, and progress in a way that is clear, current, consistent, and verifiable. This applies to Investor Relations pages just as much as to press material and restructuring explanations. Anyone who communicates contradictorily provides contradictory material to systems as well.
4. Assign Responsibility: The topic touches finance, risk management, communications, legal, and IT. If no one is responsible, no one takes care of it. Whether a dedicated role is needed or an existing department takes on the task is a leadership decision that should be made consciously.
5. Integrate into Risk Management: AIPM should not remain a communications project. Treat the results like any other risk indicator with fixed reporting channels, threshold values, and escalation levels - similar to liquidity or supply chain risks.

What AIPM Is Not

I intentionally draw a boundary here because it is crucial for the credibility of the field.

• AIPM is not manipulation. It is not about tricking systems, suppressing unpleasant facts, or forcing outcomes. Anyone who tries that will sooner or later lose trust, and the systems are continuously adapted anyway. The goal is to ensure that accurate information is reliably discoverable and correctly contextualized.
• AIPM is also not a substitute for operational performance. A company with a weak product or a poor cost structure will not become better through a polished perception. However, it can prevent a good company from being portrayed worse than it is, and it can notice earlier when the debate turns.
• And AIPM is not a settled scientific field. Open questions remain: How do models, languages, and prompts differ? How stable are the answers? Where exactly is the boundary between legitimate data curation and inadmissible influence? Who is liable when a system spreads false information about a company? I consider it a mark of professionalism to name these questions openly rather than selling certainty.

Conclusion: A Task for the C-Suite

The restructuring at BMW has solid economic reasons, and it will have noticeable consequences for many employees that deserve respect. As an expert in AI Perception Management, I draw an additional lesson from this: The next crisis - whether at BMW or elsewhere - will become visible not only in balance sheets and stock prices, but also in the summaries that decision-makers use to orient themselves.

That is why I consider it a leadership duty to observe algorithmic perception with the same sobriety as liquidity and supply chains. Not out of vanity, but out of risk awareness. Whoever does this early gains time, room to maneuver, and credibility. Whoever ignores it may find out last what everyone else has long since read.

I am interested in your perspective: Do you know what AI systems are saying about your company today? And who in your organization is responsible for it?

*Sources: manager magazin; reports from dpa, Handelsblatt, Euronews, Elektroauto-News, and others regarding BMW job cuts (as of July/August 2026).*