AI is uncovering one big and typical challenge
AI implementation is partly a knowledge audit.
In the discovery phase of a recent nonprofit project, we didn’t encounter technical constraints or legal issues. Instead we found a classic organizational problem: key-person dependency. All of us have worked with this person before.
Years of workarounds had turned one person’s undocumented judgment into part of the organization’s operating system. She couldn’t adequately explain her process because she’d never needed to. One person’s brain had become infrastructure.
Research suggests this isn’t necessarily the key person’s fault. Researchers studying what they call the “key people paradox” found a self-reinforcing cycle: key people accumulate knowledge over years, become indispensable, carry more responsibility, have difficulty relinquishing control, and eventually become too overloaded to develop successors or transfer what they know. (1)
I’m seeing versions of this in a lot of AI work. Sometimes it’s a person. Sometimes it’s the familiar Excel workbook that someone built six years ago, that everyone still uses and that three other systems now depend on. Sometimes it’s a workflow that’s partially automated but still requires a human to keep it moving.
These systems work. That’s exactly why they survive.
But they raise another question as organizations introduce AI: does the organization actually possess its knowledge in a form that anyone other than the humans carrying it around can use? A 2026 paper on human-AI decision-making describes organizational knowledge as fragmented across software systems, manual documents and tacit human expertise, much of it originally designed for human consumption. (2)
And that’s what makes changing them difficult. Introducing AI isn’t just a technology decision. It can expose years of workarounds, dependencies and undocumented decisions. Change the workflow and someone may worry about losing control, getting buried in additional work, or breaking something the organization depends on.
That makes process discovery part of AI implementation, not something that happens before it.
McKinsey’s 2026 research on AI transformation found that 48% of reported barriers to scaling AI were organizational, compared with 42% that were technical, including process redesign and unclear decision rights. (2)
The technology may actually be the easier part. But I still find the human element more rewarding. That’s where you discover how the work really gets done, and what AI actually needs to make better.
Sources:
(1). Findlay-King, L. (2026). The key people paradox: human capital vulnerability in smaller voluntary sport organisations. Voluntary Sector Review
(2)Do we have the knowledge we need? Rethinking human-AI decision-making in corporations.April 2026Anne S. R. Marx, Ricardo M. Avelino, Torbjørn Netland, Mennatallah El-Assady
(3) Rewiring Retail in Europe: The AI Imperative. June 10, 2026