The question sounds simple. Most companies answer: IT. Or R&D. Or the chief operating officer who just flew back from a conference in Berlin with a slide about an “AI-first organization” stuck in his head. None of them are right. And that's exactly the problem.
Three people to talk to a machine
For the past fifty years, talking to a computer system required a middleman. First a punched card and an operator. Then a graphical interface designed by a UX team. Then programming code — something you commission from a developer and wait two weeks for a report.
In 2012, voice assistants arrived: Siri, Alexa, Google Assistant. They promised a revolution. They didn't deliver one, because they understood commands, not context.
Today a fourth mode of interaction exists: VUI via LLM. You can ask directly: which product hasn't sold in two weeks? What were our OPEX costs in Q2 2023? Where does the delay start in the supply chain?
The system answers. No commission. No waiting. No middleman.
This isn't a change of tool. It's a change in the layer where an organization thinks.
95% of companies see no return
According to MIT research, 95% of generative AI deployments in enterprises produce no measurable financial impact. Not because the models are poor. MIT researchers point to a “learning gap” on the organization's side, not the technology's.
BCG reports that 74% of companies cannot demonstrate any real value from AI, despite active deployment.
In 2025, 42% of companies abandoned most of their AI initiatives — a sharp rise from 17% the year before. The average organization scraps 46% of proof-of-concepts before they reach production (Beam AI).
Companies aren't losing money because they bought the wrong model. They're losing it because they deployed a tool without answering one question: who in this organization decides how to use it?
The mistake every company makes the first time
AI transformation lands in IT, because IT deploys systems. Or in R&D, because that sounds innovative. Or with the COO, because “operations” is the one word that fits everything.
Every one of these decisions carries the same flaw: it treats AI as a project, not as an organizational resource.
The workforce isn't an “HR project.” ERP isn't an “IT project.” The logistics network isn't an “operations project.” They are resources — each requiring its own governance structure, its own competencies, its own line of accountability.
AI is exactly the same. A sales agent handling inquiries. An analytics agent reading data from the WMS. A communications agent processing complaints. Each has its own skill set, its own context, its own constraints. Someone has to manage this.
Not as a project. As a division.
Delegation is the new operational competency
For the past decade, “digital transformation” meant one thing: deploy an ERP, connect the WMS, automate invoicing. A hard process, coded once, running for years.
AI changes that model. The system doesn't execute one action. It understands intent and selects the response to fit the context. That demands a capability organizations have never needed before: the precise delegation of tasks to machines.
A sales rep who can't formulate a query to an AI system loses the same edge as a sales rep in 2005 who couldn't use a CRM.
An accountant who uses AI only to categorize invoices is leaving tax optimizations on the table — ones the system could surface in four seconds.
A customer service team that deploys a chatbot to answer FAQs isn't building relationships. It's cutting its own people off from the conversations where those relationships get built.
Companies that achieve real returns from AI invest 70% of their resources in people and processes, and only 10% in algorithms (Boston Consulting Group). The rest of the market does it the other way around.