L1
Transferable Knowledge
Any trained person can do this across companies and industries. Portable, not simple.
Writing an executive brief. Running a structured interview. Analysing a competitive landscape. Sequencing project dependencies.
Automate
Audit Your Knowledge
Four layers of knowledge work.
A diagnostic for seeing where your knowledge sits. So 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 Foundations.
The Audit pillar in 1:21. Prefer to read? It’s all below. ↓
Commodity / PortableUnique / Defensible
L2
Domain Knowledge
Expertise built over years in a specific field or profession.
Healthcare procurement compliance. Structural engineering judgement. Specialised tax law. Clinical diagnosis.
Augment
L3
Organisational Knowledge
Company-specific knowledge — your competitive advantage. It comes in two forms.
L3a · Documented
Written down. AI can learn it if you feed it.
Internal processes, pricing models, project playbooks, client relationship history.
L3b · Tacit
Not yet written down. AI can’t learn what isn’t there.
Why we lost that deal. Which stakeholders actually decide. The unwritten rules.
The Codification Gap
Teach
Extract
Most organisations assume their people work at Layer 3.
How much of their time is actually spent at Layer 1?
The classification test — who could do it?
L1
Anyone qualified, from anywhere
L2
Anyone in this field or profession
L3a
Anyone here who reads the docs
L3b
Only specific people here
Boundary
If you can’t tell — don’t force it. Pull it aside, name the split, and resolve it in discussion. Boundary cases are normal outputs of a real audit — great information about your organisation.
The split between Layer 1 and Layer 2
The test: Who could execute this without asking anyone in your field for context?
Any qualified professional, from anywhere → it’s Layer 1. The method is portable. Industry context isn’t required to do it well.
Only someone with years in this field → it’s Layer 2. The industry time is the expertise — regulatory pattern-matching, sector-specific judgement, accumulated stakeholder context.
Worked examples — same method, different classification
Financial analysis
Layer 1 — general
Discounted cash-flow models, variance analysis, scenario modelling.
Layer 2 — specialised
Specialist tax planning across multiple jurisdictions.
Writing a report
Layer 1 — general
A stakeholder briefing, an executive summary, a competitive analysis.
Layer 2 — specialised
A regulatory submission in a licensed industry.
Structured interviewing
Layer 1 — general
Hiring interviews, user research, stakeholder discovery.
Layer 2 — specialised
A legal deposition or expert-witness interview.
Same underlying method. The L2 versions require years of industry-specific context that can’t be compressed into onboarding. That accumulation is what makes it Layer 2.
Layer 1 commoditises first. Generic methods are the easiest for AI to replicate. Layer 2 follows, and only as models absorb enough industry-specific training data.
The split inside Layer 3
The test: Could a competent insider use the written material alone to make the same call reliably?
Yes → the documentation does the work. AI can learn it, new people can apply it. That’s Layer 3a.
No → the doc looks complete but the reasoning isn’t really in it. The call still depends on specific people’s judgement. That’s Layer 3b. It’s the layer AI can’t copy.
Worked examples — same activity, different classification
Pricing & discount decisions
Layer 3a — documented
Published price list, discount-authority matrix, approval thresholds, quarterly revenue targets.
Layer 3b — judgement
When to concede on price versus hold the line. Which customers will walk. Which are signalling churn. When a dropped margin wins loyalty and when it signals weakness.
Vendor & supplier selection
Layer 3a — documented
RFP scoring matrix, approval workflow, contract templates, reference-check protocol.
Layer 3b — judgement
Which vendors deliver versus which impress in pitches. Which sales teams over-promise. Relationship history that isn’t in the procurement record.
Hiring decisions
Layer 3a — documented
Role specification, interview rubric, competency framework, reference-check template.
Layer 3b — judgement
Reading cultural fit. When a credentialed candidate is the wrong bet for this team. Which managers can handle which personalities. When the borderline candidate is the right long-term choice.
The L3a versions are what most organisations already document. The L3b versions are what determines whether the documented procedure produces the right answer in a specific case.
The L3a / L3b line shows up in AI deployments
Brynjolfsson’s 2023 study of 5,179 customer-support agents found AI assistance produced around a third more productivity for novices and minimal gain for experts. The mechanism: AI extracted the codifiable patterns from experts’ chat logs and made them available to novices as L3a. Experts didn’t gain because the captured slice was already in their practice. The unwritten judgement that couldn’t be extracted stayed with the experts.
Dratsch et al. (2023) showed the same shape in radiology: experienced radiologists’ diagnostic accuracy fell from 82.3% to 45.5% when AI suggestions were wrong — they trusted the AI past the point where their own judgement should have overridden it. The L3b judgement is still in the experts; what changes is how easily it gets overridden.
The pattern is measurable, not theoretical. Every AI-at-scale deployment eventually hits the L3a / L3b line. Organisations that know where their line sits can deploy with confidence. Organisations that don’t are introducing unknown risk.
Why the test sits here: “Can it be written down?” is too loose. Anything can be written down badly. The reproducibility test closes the loophole. If the documentation isn’t sufficient on its own to produce the same decision, the knowledge is still tacit regardless of how much documentation exists around it.
Watch for hollow codification
Documentation without the reasoning that produced it can fool the audit. What looks like L3a — written down, retrievable — is still L3b if the written material can’t reproduce the decision on its own. We call this Hollow Codification. Structure alone isn’t sufficient. When you classify, preserve the ‘why’, not just the ‘what’.
When you can’t classify cleanly
Don’t force it. Pull the case aside, name the split, and resolve it in discussion. Boundary cases are normal outputs of a real audit. Great information about your organisation.
The framework in practice
Applied to policy and information in any regulated industry. How the four layers show up in the work.
| Layer | What lives here |
|---|---|
| L1 | Broad policies — national, state, council. Apply to any project, any industry. |
| L2 | Industry-specific policies — financial services prudential rules, pharmaceutical manufacturing standards, construction safety codes. |
| L3a | This organisation’s documented policy interpretations and application rules. |
| L3b | How this organisation applies policies day-to-day. The unwritten “what matters to them.” |
One decision, four layers — traced through
A pricing decision on a large enterprise deal — the kind every sales organisation makes every quarter — moves through all four layers before the right number is reached. This is the shape of real work.
L1
Commercial negotiation craft. Discount trading, concession sequencing, terms, walk-away discipline. Any qualified salesperson carries these from one company to the next.
L2
Enterprise pricing economics. CAC payback, gross-margin floors, multi-year discount mechanics, how ARR compounds, what a precedent-setting concession costs across the rest of the pipeline. Requires sector experience to read correctly.
L3a
This company’s discount matrix. Approval thresholds by deal size, named deal-desk approvers, standard concession catalogue, margin floors by segment, terms that require CFO sign-off. Written down. Retrievable.
L3b
How the regional VP actually decides on the borderline deal. Which customer will expand from a modest landing deal. When to trade price for a reference. How to read whether the CFO opposite is bluffing. Which concessions quietly set precedent the rest of the pipeline will inherit. 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. Pillar 1 tells you which is which. Pillar 2 activates each layer by deploying AI where it fits. Pillar 3 asks how the shape of your work changes when that line moves.
What the audit gives you
Once the four layers are visible, the next AI decision goes from intuition to three questions you can answer.
Which layer is this for?
What does it need?
Where will the value compound?
Next in the chain
Continue to Activate →Apply the lens to your architecture. See where each layer sits, and where AI deploys against it.
What was always valuable — organisational hygiene — AI just made essential.
The five forcing functions — disciplines AI made essential →The four layers are structural. What moves is the line between them. The pace of model advancement keeps accelerating. Work that needed a specialist two years ago runs through a base model today. The Layer 1 / Layer 2 line is not where it was. Your Layer 3a and 3b stay yours. Pillar 3 asks where your defensible layer sits now, and what to do with it.
Start a conversation
A question about the framework, or thinking about an engagement for your organisation? Send a note and you’ll hear back within a day or two.