# Why Even OpenAI Is Hiring Consultants, and What That Means for Your AI Strategy > OpenAI is investing four billion dollars in its own consulting subsidiary, while Anthropic is launching a parallel 1.5-billion unit. What sounds like industry news is in truth a confirmation of what we have been seeing in consulting for years: carrying AI into an organization is accompaniment work, not a license purchase. *Published: 2026-05-28* *Source: https://vellmerk.ai/en/blog/warum-openai-berater-einstellt* In May 2026, [OpenAI did something](https://openai.com/index/openai-launches-the-deployment-company/) that at first glance looks like a contradiction of its own business idea: it founded a consulting company. Four billion dollars in seed capital, 19 investors, among them Bain Capital, TPG, Goldman Sachs, plus around 150 so-called Forward Deployed Engineers who came on board immediately through the acquisition of the British consultancy Tomoro. A few days later, [Anthropic followed with its own 1.5-billion consulting unit](https://www.anthropic.com/news/enterprise-ai-services-company) together with Blackstone and Hellman & Friedman. And [PwC is certifying 30,000 of its employees on Claude at the same time](https://www.pwc.com/us/en/about-us/newsroom/press-releases/pwc-anthropic-ai-native-finance-life-sciences-enterprise-agents.html). Anyone who lays these three pieces of news side by side sees a pattern. The makers of the most powerful AI models in the world are now publicly admitting that their products alone are not enough. They are building consulting organizations, to the tune of billions. Vellmerk.ai observes this development from daily consulting practice and sees in it not a surprise but a confirmation. ## An Industry Confirms What It Long Denied This approach does not actually fit the industry's self-image. A tech company builds models, licenses them, and scales through software margins. Consulting is classically the opposite model: labor-intensive, people-heavy, hard to scale. If OpenAI nonetheless buys its way into this business with four billion dollars, then for only one reason: in reality, the models do not deliver the value they promise on paper. A [study by the NANDA initiative at MIT](https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/) has supplied the uncomfortable figure: **95 percent of all enterprise AI pilots** generate no measurable business value, despite 30 to 40 billion dollars in investment worldwide. McKinsey reaches a similar conclusion: more than 80 percent of organizations report no EBIT effect from their use of AI. The [Bitkom Study 2026](https://www.bitkom.org/sites/main/files/2026-02/bitkom-studienbericht-ki.pdf) shows the same for the domestic market: one third of companies using AI find the technology more expensive than expected, and almost one in five has already cut jobs as a result of this disappointment. These figures are readily used as proof that AI is overrated. That is the wrong conclusion. They show something else: the technology works. What does not work is the translation into the organization. ## What a Forward Deployed Engineer Actually Does The term was [coined by Palantir in the early 2010s](https://fde.academy/blog/how-palantir-invented-the-forward-deployed-engineer-model). Back then, Palantir's clients, intelligence agencies and defense authorities, worked in environments where classic software delivery did not function. Requirements were secret, systems highly specific, users not reachable in open workshops. Palantir responded with a new role model: engineers who do not build software in their own office, but work embedded on site with the client. Who pick up the problem themselves from step one, and remain responsible for the result down to the last production error. This stopgap turned into a business model. Today, OpenAI, Anthropic, Google, and nearly every serious AI company copy the pattern. A [market analysis by Pragmatic Engineer](https://newsletter.pragmaticengineer.com/p/forward-deployed-engineers) shows that job postings for Forward Deployed Engineers rose by more than 800 percent in 2025. It is the most sought-after role in the AI job market. > "When the maker of the most powerful AI in the world concludes that it has to hire consultants for its product to take effect, that should give pause to anyone who is currently buying licenses.", Thorsten Vellmerk ## What OpenAI Is Building With Billions, We Have Been Doing in Consulting for Years The engagements of Vellmerk.ai are rarely quarterly projects. From multiple client projects and over 20 years of practice, Vellmerk.ai accompanies companies over years. We know their processes, their data landscape, their strengths, and their blind spots. When we tackle an AI initiative together, it does not happen from a standing start, but on the basis of a relationship that has grown over months or years. We rebuild with our clients piece by piece: one process, then the next, then the architecture, then the enablement. What becomes visible at each step feeds into the next. This very continuity is the decisive lever. An AI initiative is not a project with a beginning and an end. It is a sequence: assessment, first lever, experience, second lever, scaling, third lever. Anyone who tries to press this into a quarter-precise supplier model will fail. Anyone who understands it as accompaniment, as a patient, continuous rebuild, has a chance. The [Klarna case](https://www.entrepreneur.com/business-news/klarna-ceo-reverses-course-by-hiring-more-humans-not-ai/491396) illustrates this well. The Swedish buy-now-pay-later provider had announced with great publicity in 2024 that AI had replaced the work of around 700 customer service employees. A year later, the CEO had to concede that service quality had declined. The company brought human agents back and now works in a hybrid model, AI for routine cases, people for escalations. Had Klarna set up the transformation from the start as an accompanied process rather than as a headline, the story would be a different one today. > "We don't do projects with a beginning and an end. We accompany organizations over years. And that is exactly how we see the things that don't become visible in a two-month pilot project.", Thorsten Vellmerk ## Four Things That Must All Be Right at Once in Every AI Initiative After years in this accompaniment work, four recurring levers have crystallized. They are not a model, but an observation: the things that must all be right at once in every AI initiative, and that in most initiatives are developed only individually. **Processes first, tools second. **The most common mistake is also the most intuitive one: a tool sounds exciting, so people ask where it might fit. The result is usually an efficiency variant of the prior state. A flawed approval workflow accelerated with AI remains a flawed approval workflow, only faster. The right order is an honest process analysis, and only then the tool question. **Onto the greenfield with courage. **The most value-creating AI workflows of the past 18 months did not exist two years ago. They are not automation of legacy, they are new possibilities. Anyone who only automates what is already there leaves the larger part of the potential untapped. The question "How would we build this process today if we were starting from scratch?" must be asked explicitly, otherwise it is implicitly answered no. That is a more expensive answer than most companies want to admit. **Architecture before license. **Between a standard license, a configured assistant, and an orchestrated workflow lie factor-of-10 differences in total cost and in time-to-value. The cheapest license is rarely the most economical solution, and the most expensive architecture is rarely the most effective. This decision does not belong in procurement. It belongs in an architecture step that considers business goal, data situation, and organizational maturity at the same time. **Enable instead of roll out. **Employees must be enabled to think differently with AI, not instructed to use a tool. A 60-minute mandatory training may satisfy formal requirements, but it does not change anyone's working day. Effective enablement means use-case workshops within one's own team, protected spaces for experimentation, and institutionalized learning formats. It is weeks and months of work, and the investment that, in our observation, is most strongly underestimated. ## What This Means for You If your company is buying licenses for AI tools today, Microsoft 365 Copilot, Claude for Business, an industry-specific AI, there is nothing wrong with that. But it is also nothing complete. The license alone creates no value. What creates value is the accompaniment work behind it: which processes fit, which must be rethought, which architecture holds, and how people grow into the new way of working. A company can build this work internally. That is not wrong, but according to the MIT study it is significantly more expensive and less often successful than accompaniment by specialized partners. It can be delivered by a large strategy consultancy, also not wrong, but rarely a good fit when the budget is below the triple-digit millions and the organization is not running ten initiatives at once. Or it can be delivered by a partner who knows the mid-market and public administration in the DACH region, who is small and independent enough for genuine architecture decisions, and who accompanies over years instead of delivering and leaving. That is exactly the kind of work Vellmerk.ai does every day. > "Carrying AI into an organization is not a delivery. It is an accompaniment. Anyone who has understood that already holds the first lever in their hand.", Thorsten Vellmerk ## Conclusion The four billion dollars that OpenAI is now investing in consulting are not industry trivia. They are the public confirmation of what consultants have experienced in practice for years: the bottleneck is not the model. It is the translation into the organization. And this translation is a craft, not a product. When you think about your next AI initiative, do not ask yourself only which tool you should buy. Ask yourself who accompanies you along the way, over months and years, not just over the next quarter. **Get in touch with Vellmerk.ai**, we take the time to understand your starting situation before we talk about solutions. ## Sources & Further Reading **OpenAI Deployment Company (May 2026), **[openai.com/index/openai-launches-the-deployment-company](https://openai.com/index/openai-launches-the-deployment-company/). Primary source on the founding, capital structure, and partner list. **Anthropic Enterprise Services Company (May 2026), **[anthropic.com/news/enterprise-ai-services-company](https://www.anthropic.com/news/enterprise-ai-services-company). Announcement of the parallel 1.5-billion initiative with Blackstone, Goldman Sachs, and Hellman & Friedman. **PwC and Anthropic (May 2026), **[pwc.com / Newsroom](https://www.pwc.com/us/en/about-us/newsroom/press-releases/pwc-anthropic-ai-native-finance-life-sciences-enterprise-agents.html). 30,000 Claude-certified employees, a joint Center of Excellence. **MIT NANDA. The GenAI Divide: State of AI in Business 2025, **[summary in Fortune](https://fortune.com/2025/08/18/mit-report-95-percent-generative-ai-pilots-at-companies-failing-cfo/). Source of the 95 percent failure rate, based on 150 interviews, 350 employee surveys, and 300 documented implementations. **McKinsey, State of AI Trust 2026, **[mckinsey.com](https://www.mckinsey.com/capabilities/tech-and-ai/our-insights/tech-forward/state-of-ai-trust-in-2026-shifting-to-the-agentic-era). EBIT effects and the development of trust in the agentic AI phase. **Bitkom, Artificial Intelligence in Germany, Study Report 2026, **[bitkom.org (PDF)](https://www.bitkom.org/sites/main/files/2026-02/bitkom-studienbericht-ki.pdf). Adoption, barriers, and profitability experiences in the German mid-market, based on 604 surveyed companies. **Forward Deployed Engineering, **[Palantir's origin model](https://fde.academy/blog/how-palantir-invented-the-forward-deployed-engineer-model) as well as [Pragmatic Engineer on the 2025 market dynamics](https://newsletter.pragmaticengineer.com/p/forward-deployed-engineers). **Klarna, CEO statement on the hybrid strategy (May 2025), **[Entrepreneur Magazine](https://www.entrepreneur.com/business-news/klarna-ceo-reverses-course-by-hiring-more-humans-not-ai/491396). Background on the return of human service agents after the pure AI phase. ## About Vellmerk.ai Vellmerk.ai is an AI consultancy (Danish ApS) founded by Thorsten Vellmerk. Drawing on 20+ years of process and IT experience and several years of hands-on AI consulting, Vellmerk.ai helps SMEs and public administration adopt AI in a practical, sovereign way, from strategy to local, on-premise-ready implementation. Proven across multiple client projects. [Book an initial consultation](/en/contact). ## Frequently asked questions ### Why are consultants still needed if AI is so capable? Because the model delivers the technology, not the translation into your context. The real work lies in finding the right use case, bringing processes, data and people together, and reliably taking the solution into operation. No model does that by itself, and that is exactly why the need for consulting rises with the capability of AI rather than falling. ### What does it mean that even OpenAI is hiring consultants? It shows that even the maker of the models knows: the technology alone generates no business value. It takes people who translate between use case and tool, sharpen requirements and accompany delivery. When the provider of the best models builds up consulting for that, it is a clear signal to everyone else. ### Does AI replace strategy consulting? It changes it but does not replace it. AI devalues the pure gathering and packaging of knowledge, that is, classic slide work. What gets upgraded, by contrast, is what AI cannot do: take responsibility, understand context, mediate between stakeholders and deliver execution dependably. Consulting shifts from analysis to execution. ### What does an AI consultant do that a model cannot? Identify the right use case, say honestly when AI is not the right solution, factor in data protection and governance, and take the solution from idea through to stable operation. A model answers questions; a consultant takes responsibility for the outcome.