Artificial Intelligence learns from what is documented, but much of the judgment that sustains an organization remains embedded in people’s experience (Polanyi’s Paradox).
In my previous article, I argued that Artificial Intelligence does not transform organizations: it amplifies what they already are. I want to explore that idea further, because the question that follows determines whether technology adoption delivers value or becomes a wasted investment: what, exactly, is AI going to amplify?
The answer is the organization’s knowledge. And that is where the problem begins, because very few organizations actually know where their knowledge resides.
In 1966, philosopher Michael Polanyi captured this idea in a phrase that should be discussed in every boardroom: “we know more than we can tell.” From this insight emerge two categories that are important to understand.
Explicit knowledge is knowledge that has been codified: manuals, policies, procedures, databases, contracts, dashboards. It can be copied, audited, transferred and—most importantly—fed into a system.
Tacit knowledge is knowledge that lives within people: judgment, accumulated experience, professional intuition, relationships and the ability to read context. It is neither informal knowledge nor second-class knowledge. In fact, it is often the knowledge that determines what happens. It is what enables an operations manager to recognize that a green indicator may be hiding a problem, or a negotiator to sense when it is time to stop pushing. Neither of them could necessarily articulate the rules they are following, yet both are often right.
The uncomfortable reality is that no one really knows how much tacit knowledge exists within their organization. Market intelligence firm International Data Corporation (IDC) states this plainly: there is no reliable way to measure that proportion, and this uncertainty is itself a strategic risk, because you cannot close a gap you cannot measure. In knowledge-intensive industries such as professional services, healthcare, advanced manufacturing and financial services, that proportion is almost certainly larger than management assumes.
The good news is that tacit knowledge can be transformed into explicit knowledge. The bad news is how much that process can cost.
Ikujiro Nonaka and Hirotaka Takeuchi documented a compelling example in The Knowledge-Creating Company. In the mid-1980s, Matsushita was developing a home bread-making machine but could not successfully mechanize the kneading process. Its engineers understood all the theory, yet the bread still came out poorly. Ikuko Tanaka, a member of the software development team, became an apprentice to the master baker at the Osaka International Hotel. It took her a year to understand a particular stretching motion that the baker performed instinctively but could not explain. That movement was eventually translated into an engineering specification—internal ribs inside the machine—and the resulting product broke sales records during its first year.
That is what it means to externalize knowledge: one person, one year, to make a single movement explicit. It is not a two-day workshop or a repository of documents.
When that work is not done, technology operates on an incomplete map. Zillow, the U.S. real estate platform, had some of the best explicit data in its industry: prices, comparable properties and transaction histories. Using that data, it built an algorithm to buy and resell homes. What the model lacked was the local judgment of real estate professionals—the understanding of why one block sells while the next one does not. My interpretation is that buying a home was never purely a data-driven decision: it is also a decision built on trust, and trust is created between people. In November 2021, Zillow shut down the business after losing $422 million in a single quarter and cutting approximately 25% of its workforce.
The consequences are already being measured. A 2025 report by MIT Media Lab’s Project NANDA reviewed more than 300 enterprise AI initiatives and concluded that 95% of organizations were seeing no return on their investment in generative AI. Its conclusion was not that the models were inadequate, but that the barriers were organizational rather than technological. IDC reaches a similar conclusion from another direction: 89% of organizations acknowledge some form of data-quality problem, and more than half of AI initiatives stall after the pilot stage.
That is why I continue to insist on the sequence. Before feeding data into any innovation initiative, organizations need to understand where that data comes from and what judgment has been left out of it. Identifying which knowledge is critical to the business, who possesses it, how it is used and how it is transferred is not an optional preliminary step: it is a prerequisite for generating a return on the investment. Without it, organizations automate incomplete processes, digitize bad practices with impeccable efficiency and build solutions that the people running the business do not recognize as their own. The outcome is predictable: employees either manually verify what the system recommends or quietly stop using it altogether.
And this work does not belong solely to the technology department. I will continue to emphasize the following:
It is strategy, because it defines which knowledge sustains the business model.
It is culture, because where knowledge is a source of power, people will not share it simply because they are told to.
It is process, because the departure of a key employee should not become an operational risk event.
And it is governance, perhaps the element organizations underestimate the most, because someone must ultimately be accountable for what gets documented, what gets automated and what should continue to depend on human judgment.
Costa Rica has talent, openness and a national strategy underway that provides us with an overall direction. But what we need to work on inside our own organizations is something different: having clarity about what we know, what has been documented and, above all, what remains intangible because it lives in the minds of our people.
Before asking whether we and our organizations are ready to adopt Artificial Intelligence, perhaps the more interesting question is whether we truly know what we claim to know. Because an organization that does not know this is hardly adopting AI: it is delegating decisions based on an edited version of itself.
If you found this article interesting, I invite you to read and learn more about Polanyi’s Paradox:
https://handwiki.org/wiki/Philosophy:Polanyi%E2%80%99s_paradox
Sources and References
IDC. (2026). The Knowledge Your AI May Never Have. Ewa Zborowska. https://www.idc.com/resource-center/blog/the-knowledge-your-ai-may-never-have/
MIT Project NANDA. (2025). The GenAI Divide: State of AI in Business 2025. Challapally, A., Pease, C., Raskar, R., & Chari, P. https://mlq.ai/media/quarterly_decks/v0.1_State_of_AI_in_Business_2025_Report.pdf
Nonaka, I., & Takeuchi, H. (1995). The Knowledge-Creating Company: How Japanese Companies Create the Dynamics of Innovation. Oxford University Press.
Polanyi, M. (1966). The Tacit Dimension. University of Chicago Press.
Zillow Group. (2021). Zillow Group Reports Third-Quarter 2021 Financial Results & Shares Plan to Wind Down Zillow Offers Operations. https://investors.zillowgroup.com/investors/news-and-events/news/news-details/2021/Zillow-Group-Reports-Third-Quarter-2021-Financial-Results--Shares-Plan-to-Wind-Down-Zillow-Offers-Operations/default.aspx
Published in La República newspaper on August 10, 2026.👈 (spanish version)