Audit your knowledge.
Four layers of knowledge work. A way to see where your knowledge actually sits. From there you can decide what moves to AI, what stays with your people, and where your real advantage is.
Grounded in sixty years of labour economics and knowledge-management theory. See Origins →
Sorted by who could do each part.
Your knowledge has structure.
Read it from the ground up. The base layers are portable work AI copies first. The summit is the work that’s hardest to copy, where your people make the difference. Each layer calls for a different response.
Hard to copy
Portable
Organisational Knowledge
Specific to your company. Your competitive advantage, in two forms.Tacit
Not yet written down. AI can't learn what isn't there.Documented
Written down. AI can learn it if you feed it.Industry Knowledge
Travels within one industry or jurisdiction, not across them.Transferable Knowledge
General work anyone qualified can do. Portable, not simple.The line where knowledge stops being written down.
Below it, AI can use what you’ve captured. Above it is what your people just know and never wrote down, and AI can’t touch that until someone writes it down. It’s the same edge where the reasoning stopped: the work above the line is yours alone.
Below the line, it’s captured. Above it, it stays with your people.
Most organisations assume their people work at Layer 3.How much of their time is actually spent at Layer 1?
General method, or industry time?
The test: who could do this without asking anyone in your industry for help? Any qualified professional → L1. Only someone with years in the industry → L2. Same method, different classification.
Discounted cash-flow models, variance analysis, scenario modelling.
Specialist tax planning across multiple jurisdictions.
A stakeholder briefing, an executive summary, a competitive analysis.
A regulatory submission in a licensed industry.
Hiring interviews, user research, stakeholder discovery.
A legal deposition or expert-witness interview.
Layer 1 goes first. Generic methods are the easiest for AI to copy. Layer 2 follows later, once models have seen enough industry-specific data.
Does the document do the work?
The test: could a competent insider use the written material alone to make the same call reliably? Yes → L3a, AI can learn it. No → L3b, the judgement still depends on specific people. That’s the layer AI can’t copy.
Published price list, discount-authority matrix, approval thresholds, revenue targets.
When to concede on price versus hold the line. Which customers will walk. When a dropped margin wins loyalty, and when it signals weakness.
RFP scoring matrix, approval workflow, contract templates, reference-check protocol.
Which vendors deliver versus impress in pitches. Which sales teams over-promise. Relationship history that isn't in the procurement record.
Role specification, interview rubric, competency framework, reference-check template.
Reading cultural fit. When a credentialed candidate is the wrong bet for this team. When the borderline candidate is the right long-term choice.
Watch for hollow codification
The trapWhat looks like L3a, written down and on file, is still L3b if the words alone can’t produce the same decision. Having it documented isn’t enough. When you classify, keep the why, not just the what.
Every large AI rollout eventually hits the L3a / L3b line.
Brynjolfsson’s study of 5,179 support agents: AI lifted novices, barely moved experts. It pulled the L3a patterns that could be written down out of the experts’ chat logs. The unwritten judgement stayed with the experts.
Radiologists’ accuracy fell when the AI was wrong. They trusted it past the point their own judgement should have taken over. The L3b judgement is still there. What changes is how easily the AI overrides it.
The pattern is measurable, not theoretical. Organisations that know where their line sits can roll out AI with confidence. Those that don’t are taking on risk they can’t see.
A pricing call, traced through.
One enterprise deal moves through all four layers before you get to the right number. This is the shape of real work.
Commercial negotiation craft
Discount trading, concession sequencing, walk-away discipline. Any qualified salesperson carries these between companies.
Enterprise pricing economics
CAC payback, margin floors, what a precedent-setting concession costs the pipeline. Requires sector experience to read.
This company's discount matrix
Approval thresholds by deal size, named deal-desk approvers, terms requiring CFO sign-off. Written down. On file.
How the VP actually decides
Which customer will expand from a small landing deal. When to trade price for a reference. Which concessions quietly set precedent. Unwritten. Held in the VP.
AI can do the top two layers today. It can read the third if you feed it the right documents. The fourth is where the work stays human.
From intuition to three answerable questions.
Which layer is this for?
What does it need?
Where will the value build over time?
Continue to Activate.
Take the four layers to your own systems. See where each one sits, and where AI fits.