A Diagnostic Framework for Generative AI in Regulated, Knowledge-Intensive Environments

by Leo Alexandru, MBA

In my role as Head of Tax Technology and Innovation for a Big 4 company, I spent countless days (and nights) trying to figure out why generative AI adoption was not working the way everyone expected it to. The puzzle was not that people were resisting the technology. They were actually using it. They liked it for drafting emails, summarizing documents, and doing quick research. The puzzle was that despite this enthusiasm, adoption remained shallow. So, when I had to choose a topic for my MBA dissertation, I had no doubt it had to be around understanding why this happens.

What Was Known, and Where It Fell Short

Anyone who studies technology adoption knows there are a few successful models, like the Technology Acceptance Model (TAM), the Unified Theory of Acceptance and Use of Technology (UTAUT) or the Technology-Organization-Environment (TOE) framework.
These are solid, well-validated models. Over decades, they have helped organizations adopt ERP systems, CRM platforms, cloud infrastructure, and dozens of other technologies. But they all share a common assumption: the technology in question behaves predictably. You put in an input, you get a reliable output. The system does what it is designed to do.
Generative AI breaks that assumption. It is probabilistic, not deterministic. It can produce something brilliant and something confidently wrong within the same conversation. In a domain like tax advisory or legal compliance, where a single factual error can trigger regulatory consequences, reputational damage, or financial loss, this changes everything.
The question shifts from “Is it useful?” or “Is it easy to use?” to “Can I trust it enough to stake my professional reputation on its output?”
None of the classic models were designed to answer that question.

A Framework Built for This Moment

Through a combination of literature review and qualitative research in the field, I developed a diagnostic framework specifically designed for GenAI adoption in knowledge-intensive, regulated environments. It builds on the strongest elements of existing models while adding three dimensions that the traditional approaches miss entirely.

The framework has five dimensions.

1. Contextual Lens, borrowed from the TOE framework. This is about understanding the environment in which adoption happens: the competitive landscape, the regulatory climate, and the regional economic and cultural context. For a firm operating across Central and Eastern Europe, for example, the drivers might be very different from those in Western Europe or the US. In some countries, demanding regulatory environments have actually pushed firms to be more digitally mature, creating an unexpected advantage. In others, risk-averse organizational cultures and limited capital slow things down. 
2. Facilitating Conditions, adapted from UTAUT. This covers the organizational infrastructure that either enables or blocks adoption: leadership support, available resources, dedicated time for experimentation, training programs, and IT systems. But the research shows that passive or merely declarative leadership support is not enough. If the CEO says AI matters but middle managers do not allocate time or resources for it, the message dies somewhere in the hierarchy. The conditions have to be real, not just announced.
3. Trust and Reliability, and this is where the framework departs most significantly from tradition. In a domain where outputs must be accurate, defensible, and compliant, the tendency of large language models to “hallucinate” is not a minor inconvenience. It is a fundamental obstacle. This dimension forces organizations to honestly assess how much they can rely on AI outputs today, what validation processes are needed, and how to build institutional confidence incrementally rather than expecting blind faith from professionals whose careers depend on being right.
4. Ethical and Regulatory Governance. None of the classic models adequately address the ethical risks of AI: data privacy, transparency, bias, and the emerging regulatory landscape, including the EU AI Act and GDPR constraints. For professional services firms, this determines what data can be used, how outputs must be documented, and what level of human oversight is required. Any adoption plan that treats governance as an afterthought is building on sand.
5. Human-Centric Factors. This goes beyond the standard “change resistance” narrative. It includes the fear of professional obsolescence, which in fields like tax preparation is empirically justified by research showing automation susceptibility above 90%. It includes the cultural stigma some professionals feel about admitting they use AI, as if it somehow diminishes their expertise. It includes the skills gap between those who use the tools and those who must supervise the outputs. And it includes the unrealistic expectations of top leadership that leads to disappointment when the technology fails to deliver perfection on the first try.

The Aha Moment

The deeper I went, the clearer it became that the reason most GenAI adoption initiatives underperform has nothing to do with the technology itself. The tools are good enough. They are getting better every quarter. The real problem is that organizations are applying a playbook designed for predictable, rule-based systems to a technology that is fundamentally unpredictable.

They are measuring the wrong things, addressing the wrong barriers, and ignoring the dimensions that actually determine whether professionals will move from casual experimentation to genuine integration.

GenAI is not just a faster spreadsheet or a smarter search engine. It is a shift that requires a new category of thinking about adoption. And the organizations that will lead in this space are not the ones that rush to deploy the latest model but those that take the time to diagnose their own readiness across all five dimensions, honestly and without shortcuts.

How Can a Manager Use This?

If you lead a team, a practice, or a business unit in any knowledge-intensive sector, here is a practical way to use this framework. Take each of the five dimensions and ask yourself a simple diagnostic question.

1. Do we understand the specific regulatory, cultural, and competitive dynamics of the markets we operate in, and are we adapting our AI strategy to those realities, or are we importing a generic plan from headquarters?
2. Beyond the corporate messaging, have we actually allocated dedicated time, budget, and resources for our people to learn and experiment with AI? Or are we asking overworked professionals to figure it out on their own?
3. Do we have clear, honest guidance on what AI can and cannot reliably do in our specific domain? Have we built validation processes that professionals actually trust, or are we asking them to “just try it” and hoping for the best?
4. Do our people know exactly what data they can use, in which tools, and under what conditions? Or are they operating in a grey zone where caution leads to paralysis and boldness leads to risk?
5. Are we addressing the real fears and skills gaps, the ones people do not bring up in town halls? Have we created safe spaces for experimentation where failure is a learning opportunity and not a career risk?

If you can answer all five with confidence, you are ahead of most organizations. If you find gaps, you have found exactly where to focus before investing another dollar in AI tools.

This article is based on the diagnostic framework Leo developed as part of his MBA consultancy project at BISM, focused on generative AI adoption in professional services across Central Europe.

Leo Alexandru is a technology and business leader with over 18 years of experience building and scaling tech-driven organizations. He leads Tax Technology and Innovation for Deloitte Romania, focusing on AI adoption, digital transformation, and operating models for complex, regulated environments. He teaches Digital Transformation at the university level, writes “The Antifragile Intelligence” newsletter on Substack, and is building courses on AI literacy and strategy for leaders navigating the intersection of technology and business.