NAKEDAI FRAMEWORK
The AI Integrations Matrix.
A board-level map of where AI sits, what it touches, who owns it, and how much governance it needs. Use it to calibrate the size of the pause to the size of the decision.
Map the decision before you size the governance
Use this framework when an AI proposal is on the table and the discussion has moved to platforms, suppliers or pilots before anyone has established what kind of decision this actually is.
The same technology can be a minor convenience for one person or a system the organisation depends on. The interface may look identical in both cases. The decision is not the same, and neither is the governance it needs.
Two questions place any AI use on the matrix. How much does it influence a material decision, and how deeply is it integrated into the way the organisation works?
Decision impact
Decision impact is not about how capable the output is. It is about what happens because of it. Ask a simple question: if this output were wrong, and nobody caught it, what would follow?
Low: The consequence stops with the person using it, or with work that is checked again before it goes anywhere. A first draft that will be rewritten. A summary that helps someone read faster. If it is wrong, someone notices and fixes it.
High: The output shapes a choice that affects money, people, obligations or reputation. A briefing a senior decision-maker acts on. An assessment that determines how someone is treated. Work that goes to a client. If it is wrong and nobody catches it, something happens that cannot easily be undone.
The test is not the size of the task. It is the distance between the output and a consequence.
Depth of Integration
Depth of integration is about how far the AI has become part of how the work runs, rather than something a person chooses to use.
Ask: if this were switched off tomorrow morning, what would break?
Ad hoc or informal: The AI sits beside the work. One person uses it because they find it helpful. It draws on material they give it. If it stopped, they would go back to how they worked before.
Embedded or business-critical: The AI sits inside the work. It connects to live systems and organisational data. It runs repeatedly, often without anyone deciding each time. Other processes now assume its output exists. If it stopped, the work would stop with it.
The test is not how sophisticated the connection is. It is how much of the organisation depends on it.
Why both questions are needed
Either one on its own gives a misleading answer.
Something can be deeply embedded and still low impact: a system that reconciles routine records all day, where each individual output barely matters. The risk there is scale. A small error repeats many times before anyone sees the pattern.
Something can be entirely ad hoc and very high impact: one person using a chat tool, occasionally, to prepare advice that a client acts on. The risk there is not scale. It is that nobody thinks of it as a system at all, so nobody governs it.
The two questions together show which kind of risk the organisation is holding.
The four positions
1. Personal productivity
Low impact, ad hoc integration.
An individual uses AI to work faster and think better: drafting, summarising, research prompts. The output stays with them, and they are close enough to it to see when it is wrong.
Ownership sits with the individual, with manager guidance. Governance should be proportionate: acceptable-use guidance and data awareness.
2. Team workflow
High impact, ad hoc integration.
A department uses AI in repeatable workflows: legal review, HR screening, marketing production. The output no longer stays with one person. It travels into other people's work and informs what they do next.
Ownership moves to a function lead. This position needs policy, review and consistency, because one person's unchecked output becomes another person's starting assumption.
3. Operational process
Low impact, embedded integration.
AI becomes part of how the business runs: customer service, finance checks, procurement analysis. Each individual output may be minor. The volume is not.
Ownership sits with operations or the process owner. This position needs controls, an audit trail and escalation, because a small error repeats at scale before anyone sees the pattern.
4. Decision support
High impact, embedded integration.
AI influences material decisions: hiring, lending, pricing, mergers and acquisitions, compliance. These decisions have real consequences for real people.
Ownership requires an executive sponsor and board accountability. This position needs human oversight, explainability and defensibility. Oversight must be real, which means the reviewer needs time, context and the authority to reject the output.
Map two positions, not one
Most organisations map where a pilot sits today and stop there. The more useful question is where success will take it.
A pilot may use copied files, a small group of users and manual review. The intended operating model may connect to live data, serve several departments and shape decisions that matter. Treating both as the same use creates a false sense of continuity. Technical capability often scales smoothly. Ownership, review capacity and risk do not.
Write down two positions before a pilot begins. Where the AI sits during the pilot, and where it will sit if the pilot succeeds. Then describe what has to be true to move between them.
That movement is the decision.
The AI you did not decide to buy
Most organisations map where a pilot sits today and stop there. The more useful question is where success will take it.
A pilot may use copied files, a small group of users and manual review. The intended operating model may connect to live data, serve several departments and shape decisions that matter. Treating both as the same use creates a false sense of continuity. Technical capability often scales smoothly. Ownership, review capacity and risk do not.
Write down two positions before a pilot begins. Where the AI sits during the pilot, and where it will sit if the pilot succeeds. Then describe what has to be true to move between them.
That movement is the decision.
The leadership check
Use these questions with your leadership team.
If an answer is vague, leave it blank and treat that as useful evidence.
Current position
Where does this AI use sit on the matrix today?
Intended position
Where will it sit if it succeeds?
Conditions to move
What must be true before it moves from one position to the other?
Ownership
Who owns it now, and who owns it then?
Embedded AI
What AI is already running inside software we pay for, and where does that sit?
Size the pause to the size of the decision
A personal drafting tool does not need a board process. A system influencing material decisions does not become safe because it has the same chat box as the drafting tool.
The purpose is not to make every AI use look large. It is to stop a large use being governed as though it were small.
If you cannot agree where this sits, or where it is heading, take a Human Pause before approval.