The most common codification mistake isn’t refusing to write things down. It is writing down the wrong thing.
Teams ask their experienced people to document how they make decisions. Those people describe a process: the inputs they look for, the steps they follow, the signals they check. The documentation looks thorough. New people read it. They apply it. And then they produce work that follows every step and still gets rejected by the expert who wrote it.
This is not a documentation quality problem. It is the specific, harder problem of codifying judgment: the one that sits underneath the easier problem of codifying process.
Earlier this week I argued that codification is the AI-era competitive advantage: the organisations building something durable are the ones giving their AI something real to check itself against. But there is a distinction inside that argument that matters. Process and judgment are not the same thing, and codifying one does not get you the other.
Process describes the normal case. Judgment governs the rest.
A process is what happens when everything goes as expected: standard inputs, familiar decision points, predictable outputs. It is possible to document a process completely and still fail to capture what makes it produce good results when edge cases arrive.
Judgment is what kicks in when the process runs out. The instinct that says this scope is too broad even though it passed every pre-defined criterion. The customer communication that is technically correct and somehow still wrong in tone. The feature that meets every requirement on paper and violates the product’s purpose in practice. Experienced people catch these things immediately. They often cannot say exactly why until someone asks.
The philosopher Michael Polanyi called this the tacit dimension: we know more than we can tell. Expert judgment accumulates through years of pattern recognition: thousands of cases, hundreds of corrections, dozens of moments where something felt off before the reason became clear. That accumulation is real and reliable. It just does not transfer through description.
Ask an expert to write down how they make a decision and they will describe what they think they do. What they actually do is harder to surface.
The standard method extracts the wrong thing
The usual codification approach (ask experienced people to document their decision criteria) produces the same result in most organisations. Criteria that describe good work in the abstract. Criteria any experienced person would agree with and that provide no real guidance when difficult cases arrive.
“Maintain quality standards.” “Prioritise customer value.” “Ensure the scope is appropriate.” All true. Universally agreed on. Useless when two experienced people look at the same case and reach different conclusions.
The problem is that you are asking experts to articulate something they have never needed to articulate before. Their judgment is operational, not declarative. They apply it constantly. They have rarely had to state it as rules transferable to someone who doesn’t share their context.
When you ask them to do this, they produce what they think a well-written standard should look like. Not what they actually use when the hard case lands.
Where expert judgment surfaces
The place where tacit judgment becomes visible is not in the description of normal work. It is in the reactions to abnormal work.
When an expert looks at something and immediately says “no, that’s wrong” before they can explain why: there is judgment in that reaction. When two experienced people look at the same case and reach different conclusions: their disagreement is diagnostic. When someone overrides the standard approach because it produced a result they know is incorrect: that override is data.
These are the moments to mine.
The right extraction method targets them directly. Instead of “how do you make this decision?”, ask: “Tell me about the last time you looked at something and immediately knew it was wrong. What did you see?” Or: “When did you last override what the process produced? What did you see that the process missed?” Or: “Describe a case where applying the standard criteria gave you a result that still felt incorrect. What was different about it?”
The answers surface what actually governs expert judgment: not the principles they espouse, but the patterns they have learned to catch. The edge cases where standard rules fail. The contextual factors that change what “good” looks like. The things experienced people check that nobody explicitly told them to check.
In one organisation I worked in, codifying a content quality standard produced twelve principles on the first pass. Every reviewer signed off on them. They described good content accurately. They did not stop bad content from passing review. It took showing reviewers specific rejected pieces and asking them to explain the rejection (“what exactly was wrong with this one?”) before the operative criteria emerged. Those criteria were more specific, more contextual, and harder to articulate than the original twelve. They were also the ones that actually worked.
The AI payoff
AI has no judgment. It has criteria.
When criteria come from the normal case (what the process is supposed to produce), they catch normal errors. When criteria come from the edges, from the exceptions and overrides and rejections, they catch what your best people would catch.
I have argued elsewhere that the organisations succeeding with AI will be the ones that build the discernment to match their generation capability. That discernment starts here: knowing that asking experts to describe their process is not the same as capturing their judgment, and having a method for getting to the harder thing.
Process documentation is the starting point. The destination is criteria specific enough to catch what your best people would reject. You get there through the exceptions.
The third post in this series covers what keeping codification alive looks like over time: The Living Standard (4 July 2026).
This post is part of The Codification Advantage series. Start with the series opener if you haven’t read it.
Leave a Reply