Approach.
AX is run as a quarterly loop, not a one-time strategy document that goes stale in six months.
Read
Patterns, not predictions.
Before designing anything, the studio reads what already shipped. Walmart productized its substrate. Visa built a trust franchise. UnionPay hit the limits of productization under regulatory mandate. These are not case studies for inspiration. They are patterns — repeatable shapes of how AI actually lives inside a company, with all its sandbox constraints intact.
Every engagement starts with a Pattern Library of 10 to 20 companies in your shape of business — sector, scale, ownership model, regulatory environment. Most strategy decks skip this step. That is why most strategy decks are wrong.
Pattern Library — 10–20 companies, your shape
Match
Most patterns won't survive your sandbox.
Once the patterns are on the table, the question is which ones survive your domain, your structure, and your sandbox — the six constraint layers (information, departments, legal, culture, history, future). A pattern that works for Visa fails at a mid-market manufacturer because the trust franchise depends on a four-party rail. A pattern that works for a US-based SaaS fails in Korea because the AI Framework Act re-frames the data handling.
The output is not a roadmap. It is a shortlist of two or three patterns that survive your sandbox — and a Stop-Doing List of patterns to abandon before they damage anything. The second list is often more valuable than the first.
Read and Match define your AI Strategy — pattern-by-pattern, not opinion-by-opinion.
Pattern Library — 10–20 companies, your shape
Build
Build is bespoke. Every person. Every department.
This is where most AX programs lose the plot. They buy tools before they have governance. They roll out one policy that fits one team and quietly damages four others. They confuse training events with habits of consumption.
Build runs in a different order:
01
Governance first
AI Council, one-page policy the CFO can sign, escalation paths. A working structure, not a fifty-page document.
02
Bespoke fitting
The pattern from Stage 02 doesn't apply identically to sales and to legal. Each person, department, and regulatory boundary gets its own configuration of what AI may do and may not.
03
Orchestrator
A single coherent layer that knows every adjustment, exception, and line. Without it, every AI tool is a separate risk surface.
04
Consumption habits
Routines that make AI use second nature. Not training events that everyone forgets in a week. The habits close the loop.
Build is the longest stage and the one that breaks programs that skip it. The other stages produce documents. This one decides whether the documents matter.
AI Operating Model (Charter · Council · Roadmap · Risk Matrix · Standards Mapping)
Loop
The cycle is the work.
Every quarter we re-run the stages with new inputs. New patterns from companies that shipped something in the last 90 days. New domain conditions — your competitors moved, your customers changed, a model got cheaper. New sandbox realities — a regulation took effect, a department reorganized, a cultural norm shifted. Read again. Match again. Update the bespoke fit. Re-teach the orchestrator. Reinforce the habits. The loop is your strategy.
Companies that treat AX as a one-time project lose to companies that treat it as an operating cadence. There is no version of AX you finish — only versions you maintain or let drift.
Quarterly Loop Review + Board Memo
Return to Stage 01 — with new inputs, every 90 days.
The loop continues
The loop continues
Four functions. Every deliverable lives in one.
The AX Loop runs the work. NIST AI RMF's four functions classify the work. Every Council agenda item, every deliverable, every quarterly review enters one of these four. Items that don't classify aren't Council items.
01
Govern
What we follow. What we forbid.
Loop stage
Read + Build (principles, charter)
Deliverable
AI Charter · Council Decision Rights
02
Map
Where we are. What's on the field.
Loop stage
Read + Match (patterns, sandbox)
Deliverable
Pattern Library · Sandbox Map · Data Tier Matrix
03
Measure
How we know it's working.
Loop stage
Match + Loop (KPIs, quarterly review)
Deliverable
Quarterly KPI Dashboard · Graduation Criteria
04
Manage
When something breaks, what happens.
Loop stage
Build + Loop (response, sunset)
Deliverable
60-min Incident Response · Escalation Paths
01
Govern
What we follow. What we forbid.
Loop stage
Read + Build (principles, charter)
Deliverable
AI Charter · Council Decision Rights
03
Measure
How we know it's working.
Loop stage
Match + Loop (KPIs, quarterly review)
Deliverable
Quarterly KPI Dashboard · Graduation Criteria
02
Map
Where we are. What's on the field.
Loop stage
Read + Match (patterns, sandbox)
Deliverable
Pattern Library · Sandbox Map · Data Tier Matrix
04
Manage
When something breaks, what happens.
Loop stage
Build + Loop (response, sunset)
Deliverable
60-min Incident Response · Escalation Paths
If a deliverable doesn't classify as Govern, Map, Measure, or Manage — it doesn't belong in the engagement.
The Two Weights Test + The Sandbox Map.
Before we start the loop, two diagnostics run first. The Two Weights Test asks whether the work even contains AX. The Sandbox Map asks whether your company has the constraints under which AX has to operate. Skip either and the loop spends a quarter on the wrong work or in the wrong place.
Diagnostic 01 — The Two Weights Test
Test 01
Information
If the work could be done by someone with a completely different background, there is nothing special to encode. If a junior could do it with a few weeks of training, still nothing special.
If yes → proceed to Test 02.
If no → this is automation, not AX.
Test 01
Information
If the work could be done by someone with a completely different background, there is nothing special to encode. If a junior could do it with a few weeks of training, still nothing special.
If yes → proceed to Test 02.
If no → this is automation, not AX.
Test 02
Can the weights be made conditional?
A machine cannot learn "works like someone experienced." It can only learn conditions. In this situation, with these signals, this person decides like this.
If yes → the loop begins.If no → tacit knowledge is trapped.
Train first.
If partial → start with the conditional fraction.
Diagnostic 02 — The Sandbox Map
Your company is a sandbox. Six layers of constraint that every AI deployment runs into — and breaks if it isn't respected. The deployment is not the hard part. Making it fit is.
YOUR COMPANY
01 — Information
What stays in.
Some information leaves freely. Some never leaves. Some has to be preserved for futures you can't see yet.
02 — Departments
Different rhythms.
Sales weekly. Finance quarterly. Legal case-by-case. One AI policy that fits all of them fits none.
03 — Legal
Region by region.
EU AI Act, Korean AI Framework Act, sector regulations. Legal in one jurisdiction, a liability in another.
04 — Culture
Team by team.
What sounds like decisiveness in one team sounds like aggression in another. AI that copies one voice damages the rest.
05 — History
Records that matter.
Why-we-did-it-that-way knowledge. Most of it isn't in your CRM. Most walks out the door if you automate it away.
06 — Future
What to preserve.
Some capabilities aren't valuable today but will be in two years. Automating them away looks like efficiency now.
DEPLOY AI HERE
Deploy AI without mapping all six layers, and the work isn't AX.
It's automation that quietly breaks the things your company already does well.
AX = the fit, not the deploy
Approach.
AX is run as a quarterly loop, not a one-time strategy document that goes stale in six months.
Read
Patterns, not predictions.
Before designing anything, the studio reads what already shipped. Walmart productized its substrate. Visa built a trust franchise. UnionPay hit the limits of productization under regulatory mandate. These are not case studies for inspiration. They are patterns — repeatable shapes of how AI actually lives inside a company, with all its sandbox constraints intact.
Every engagement starts with a Pattern Library of 10 to 20 companies in your shape of business — sector, scale, ownership model, regulatory environment. Most strategy decks skip this step. That is why most strategy decks are wrong.
Pattern Library — 10–20 companies, your shape
Match
Most patterns won't survive your sandbox.
Once the patterns are on the table, the question is which ones survive your domain, your structure, and your sandbox — the six constraint layers (information, departments, legal, culture, history, future). A pattern that works for Visa fails at a mid-market manufacturer because the trust franchise depends on a four-party rail. A pattern that works for a US-based SaaS fails in Korea because the AI Framework Act re-frames the data handling.
The output is not a roadmap. It is a shortlist of two or three patterns that survive your sandbox — and a Stop-Doing List of patterns to abandon before they damage anything. The second list is often more valuable than the first.
Read and Match define your AI Strategy — pattern-by-pattern, not opinion-by-opinion.
Pattern Library — 10–20 companies, your shape
Build
Build is bespoke. Every person. Every department.
This is where most AX programs lose the plot. They buy tools before they have governance. They roll out one policy that fits one team and quietly damages four others. They confuse training events with habits of consumption.
Build runs in a different order:
01
Governance first
AI Council, one-page policy the CFO can sign, escalation paths. A working structure, not a fifty-page document.
02
Bespoke fitting
The pattern from Stage 02 doesn't apply identically to sales and to legal. Each person, department, and regulatory boundary gets its own configuration of what AI may do and may not.
03
Orchestrator
A single coherent layer that knows every adjustment, exception, and line. Without it, every AI tool is a separate risk surface.
04
Consumption habits
Routines that make AI use second nature. Not training events that everyone forgets in a week. The habits close the loop.
Build is the longest stage and the one that breaks programs that skip it. The other stages produce documents. This one decides whether the documents matter.
AI Operating Model (Charter · Council · Roadmap · Risk Matrix · Standards Mapping)
Loop
The cycle is the work.
Every quarter we re-run the stages with new inputs. New patterns from companies that shipped something in the last 90 days. New domain conditions — your competitors moved, your customers changed, a model got cheaper. New sandbox realities — a regulation took effect, a department reorganized, a cultural norm shifted. Read again. Match again. Update the bespoke fit. Re-teach the orchestrator. Reinforce the habits. The loop is your strategy.
Companies that treat AX as a one-time project lose to companies that treat it as an operating cadence. There is no version of AX you finish — only versions you maintain or let drift.
Quarterly Loop Review + Board Memo
Return to Stage 01 — with new inputs, every 90 days.
The loop continues
Four functions. Every deliverable lives in one.
The AX Loop runs the work. NIST AI RMF's four functions classify the work. Every Council agenda item, every deliverable, every quarterly review enters one of these four. Items that don't classify aren't Council items.
01
Govern
What we follow. What we forbid.
Loop stage
Read + Build (principles, charter)
Deliverable
AI Charter · Council Decision Rights
02
Map
Where we are. What's on the field.
Loop stage
Read + Match (patterns, sandbox)
Deliverable
Pattern Library · Sandbox Map · Data Tier Matrix
03
Measure
How we know it's working.
Loop stage
Match + Loop (KPIs, quarterly review)
Deliverable
Quarterly KPI Dashboard · Graduation Criteria
04
Manage
When something breaks, what happens.
Loop stage
Build + Loop (response, sunset)
Deliverable
60-min Incident Response · Escalation Paths
If a deliverable doesn't classify as Govern, Map, Measure, or Manage — it doesn't belong in the engagement.
The Two Weights Test + The Sandbox Map.
Before we start the loop, two diagnostics run first. The Two Weights Test asks whether the work even contains AX. The Sandbox Map asks whether your company has the constraints under which AX has to operate. Skip either and the loop spends a quarter on the wrong work or in the wrong place.
Diagnostic 01 — The Two Weights Test
Test 01
Information
If the work could be done by someone with a completely different background, there is nothing special to encode. If a junior could do it with a few weeks of training, still nothing special.
If yes → proceed to Test 02.
If no → this is automation, not AX.
Test 02
Can the weights be made conditional?
A machine cannot learn "works like someone experienced." It can only learn conditions. In this situation, with these signals, this person decides like this.
If yes → the loop begins.If no → tacit knowledge is trapped.
Train first.
If partial → start with the conditional fraction.
Diagnostic 02 — The Sandbox Map
Your company is a sandbox. Six layers of constraint that every AI deployment runs into — and breaks if it isn't respected. The deployment is not the hard part. Making it fit is.
YOUR COMPANY
01 — Information
What stays in.
Some information leaves freely. Some never leaves. Some has to be preserved for futures you can't see yet.
02 — Departments
Different rhythms.
Sales weekly. Finance quarterly. Legal case-by-case. One AI policy that fits all of them fits none.
03 — Legal
Region by region.
EU AI Act, Korean AI Framework Act, sector regulations. Legal in one jurisdiction, a liability in another.
04 — Culture
Team by team.
What sounds like decisiveness in one team sounds like aggression in another. AI that copies one voice damages the rest.
05 — History
Records that matter.
Why-we-did-it-that-way knowledge. Most of it isn't in your CRM. Most walks out the door if you automate it away.
06 — Future
What to preserve.
Some capabilities aren't valuable today but will be in two years. Automating them away looks like efficiency now.
DEPLOY AI HERE
Deploy AI without mapping all six layers, and the work isn't AX.
It's automation that quietly breaks the things your company already does well.
AX = the fit, not the deploy