Why the old blueprint no longer fits
The digital transformations of the last fifteen years had one convenient trait: a lot of it was generic. Moving to the cloud, introducing a new ERP, consolidating data centers - these are topics that can be managed through a central, overarching interface function. You build a big transformation office, define standards, and push them through the organization. That is exactly why the logic of the central program emerged.
The catch: these programs were always considered difficult. In its report “Flipping the Odds of Digital Transformation Success" (2020), BCG found that 70 percent of digital transformations miss their goals - 30 percent succeed fully, 44 percent create some value but miss their targets, 26 percent deliver little to nothing. But even when they did succeed, they worked because the core was generic.
With AI, the core is not generic. AI creates value not at the infrastructure level, but at the level of the individual work process. It is not about “we are moving to the cloud", but about “how exactly does accounts receivable create its dunning runs, and what part of that can an agent take over". That is different in every department. The individual workflow becomes hugely more important.
The data clearly supports this. The MIT study “The GenAI Divide: State of AI in Business 2025" (MIT NANDA, July 2025, lead author Aditya Challapally, based on over 300 deployments, 52 interviews, 153 surveyed executives) found that 95 percent of generative AI pilots in companies deliver no measurable return - despite 30 to 40 billion dollars in investment. The reason is not the technology, but the lack of integration into real workflows. McKinsey reaches the same conclusion from the other direction in its “State of AI 2025" report: among all organizational factors studied, fundamental workflow redesign has the strongest correlation with actual EBIT impact. And only 21 percent of companies using generative AI have fundamentally redesigned even one workflow. The rest lay AI like a thin layer over existing processes - and wonder why nothing comes out the bottom.
Many small microtransformations instead of one big office
If the value lies in the individual process, then the transformation has to start there. Not one big transformation office, but many small microtransformations. Ideally one per department, aligned to the related workflows.
That sounds like more effort, but it is faster. A microtransformation asks three questions for one clearly defined area: What exactly happens here today, step by step? What can be changed or automated with which tools? And what has to be rebuilt from scratch because the existing process is simply not made for AI? These questions can be answered in weeks, not quarters - but only if someone knows the domain well enough to truly understand the process.
BCG also points in this direction. The “AI Radar 2025" describes the 10-20-70 principle. Sylvain Duranton, global head of BCG X, puts it this way: successful companies spend 70 percent of their effort on changing people, processes, and culture, 20 percent on data and technology, and only 10 percent on the algorithms. No central office can do those 70 percent for the whole organization at the same time. They happen department by department, in the actual day-to-day work.
Why freelancers are ideal for microtransformations
This is the part where, as a founder, I am naturally biased - but the logic holds even without my own interest. For microtransformations, you need people who bring two things at once: current AI expertise and the functional know-how of the respective domain. Someone who understands how a recruiting process works - and how to apply an agent to it.
Building such profiles internally takes time. The market barely offers them: according to Bitkom (press release from August 7, 2025), Germany was short about 109,000 IT specialists; 85 percent of companies complain about a shortage, 79 percent expect it to get even worse. Whoever waits for internal hiring is waiting for a market that is getting tighter.
For this task, freelancers have a structural advantage that I know well from my own time in large companies: they are less constrained by general data protection rules and the company’s lengthy procurement processes. An external expert can test a tool, build a prototype, set up a POC - without triggering a six-month vendor approval process. And that POC is worth its weight in gold. It turns an abstract discussion (“should we use AI in collections?") into something tangible that can be used internally to make the case.
The MIT numbers also argue for looking outside: according to the study, AI solutions implemented with external, learning-capable partners reach production in about 67 percent of cases - internal in-house builds only in about 33 percent, so half as often. Externals bring experience from many implementations - something an internal team simply cannot have by definition.
What this looks like in practice
Three projects we are currently supporting at FRATCH show the pattern - anonymized, but real.
At an international agency group, the microtransformation is happening in HR. The starting point was one workflow: reviewing and pre-qualifying applications across several country companies. A freelancer with HR and AI background built a POC in just a few weeks that structures resumes and matches them against role requirements. No company-wide program - one process, one department, one tangible result.
At a consulting firm, it was about internal knowledge work: pulling together proposals, references, and methods from old projects. Here too, the entry point was a single, clearly defined process where an external person could test what a tool does before the organization invested heavily.
At a large German bank, the framework is stricter - regulation and data protection leave little room. That is exactly why the micro approach works: instead of forcing a big project through compliance, a narrowly defined process in the back office is selected, one that can show under controlled conditions what is possible. The POC becomes an argument toward the internal committees.
The pattern is always the same: you do not start with a strategy slide, but with a real workflow and someone who understands it.
What this means for your next decision
If you are planning an AI transformation, the first question is not “Who leads our transformation office?", but “Which department, which process, who builds the first POC?". Look for an area with clear pain and manageable scope. Bring in someone for that area who understands AI and the domain. Let them build something tangible quickly. And repeat that, department by department.
The irony: the fastest way to AI transformation does not go through the biggest program, but through the smallest units. That is exactly why access to the right people - AI freelancers with domain knowledge, sourced quickly - often determines the speed more than any roadmap.
