Engage · Engagement
AI Operating Model & Governance Design
Close the gap between AI ambition and accountable execution.
Decide where AI authority should sit, who remains accountable, and what controls must exist before AI moves into production.
The problem
AI activity is not an AI operating model.
Many organisations now have AI pilots, an AI policy, a steering committee and a set of technology choices. What they rarely have is a coherent operating model for accountable AI execution: a deliberate answer to where AI may act, who owns what it does, and who can stop it.
The gap is not simply “AI governance”. It sits between executive ambition and operational authority, between model capability and business accountability, between risk policy and production control, and between implementation activity and decision rights. Call it the governance chasm: the distance between what the organisation intends AI to do and the structure that would let it do so accountably.
Left alone, that structure is not absent. It is inherited — from procurement decisions, cost metrics and whichever implementation partner arrived first — and it is usually discovered in the first incident.
You will recognise this if
- The CIO is held accountable for AI outcomes, while business units own the use cases.
- Data, model, risk, security and delivery responsibilities are distributed across functions that each own one part.
- Nobody has an explicit answer to who can approve, override, pause or stop an AI-enabled process, or who owns the result when it is wrong.
- Implementation partners are defining the structure while also delivering against it.
Who it is for
- CIOs and enterprise technology leaders
- Transformation executives
- Boards and executive teams sponsoring material AI programmes
For organisations moving from experimentation toward scaled or production use of AI. It is not aimed at selecting tools, architectures or models.
What the engagement answers
- Where should intelligence live in the organisation?
- Which decisions may AI advise on, prepare, recommend or execute?
- Who owns the outcome when several functions control different parts of the system?
- Who has the authority to approve, override, pause or shut down an AI-enabled process?
- Which controls should be tiered by risk rather than applied uniformly?
- How should incentives, funding and performance measures reinforce the intended model?
- What must be true before production deployment?
- Which governance responsibilities belong to management, risk, technology and the business?
What you receive
Decision-useful artefacts for the people who own the outcome. Not a generic strategy report and not a framework to file.
01
Current-state governance and accountability diagnostic
Where authority, ownership and control actually sit today, as distinct from the organisation chart and the policy.
02
AI decision-rights map
For each material decision domain: whether AI advises, prepares, recommends or executes, and who holds the decision.
03
Outcome ownership and accountability model
One accountable owner per outcome, with authority over the data, model, workflow and risk that produce it.
04
Risk-tiered governance design
Control proportionate to consequence, instead of one committee path that is too slow for low risk and too thin for high risk.
05
Approval, escalation, override and shutdown paths
Who may approve, who is told, who may override, and who may stop the system, named before go-live.
06
Production control requirements
The measures, thresholds, escalation and decommissioning conditions under which an AI-enabled process runs.
07
Operating cadence and executive governance forum design
How the model is held over time: the forums, their decision rights and the cadence that keeps them current.
08
Implementation priorities and sequencing
The decisions and changes required, in the order that makes the rest possible.
How it works
1. Frame the decision
Identify the AI ambition, the material use cases and the governance question that has to be answered.
2. Map the structure
Examine ownership, authority, incentives, funding, risk and production controls as they operate today.
3. Test the gaps
Find where accountability is split, delayed or assumed rather than held.
4. Design the model
Define decision rights, control tiers, escalation paths and the operating cadence.
5. Leave an executive blueprint
Hand over the decisions and priorities implementation must respect.
A bounded, senior-led engagement. Not a large team and not a multi-year programme.
The boundary
This is
- ✓Senior-led advisory
- ✓Operating-model and governance design
- ✓Independent challenge
- ✓Clarification of decision rights
- ✓Preparation for accountable implementation
This is not
- ✕Platform selection
- ✕Model selection
- ✕Systems integration
- ✕AI engineering
- ✕Managed services
- ✕Outsourced governance administration
- ✕A transformation PMO
- ✕A substitute for management ownership
Implementation may follow. This engagement defines the structure implementation must respect.
The thinking behind it · Perspective 3
Where Should Intelligence Live, and Who Controls It?
An AI operating model is designed or it is inherited: an organisation that does not decide, decision by decision, where intelligence lives and who controls it will have that answer set by procurement, cost metrics and its first incident.
Its core thesis
- 01Placing intelligence is allocating authority.
- 02One person must own the outcome, with authority over the system that produces it.
- 03Governance should be tiered by risk, not imposed through one committee path.
- 04Capital and scorecards influence where intelligence ends up, before architects do.
- 05Production control is the operating model at runtime.
The wider evidence
Of 25 attributes tested, the redesign of workflows had the biggest effect on an organisation’s ability to see EBIT impact from generative AI, and only 21% of respondents reported fundamentally redesigning at least some workflows. A CEO’s oversight of AI governance was among the attributes most correlated with bottom-line impact. Scaling is an organisational and operating-model change, not more pilots.
McKinsey & Company, “The state of AI: How organizations are rewiring to capture value”, March 2025 ↗Half of the 310 executives surveyed named their biggest hurdle as turning responsible AI principles into practice: scaling and automating governance, clarifying roles and responsibilities, and aligning leadership. Trust in AI is built on clear ownership and working controls, not on stated principles.
PwC, “2025 Responsible AI survey: From policy to practice”, 2025 ↗Cited as independent evidence. Neither firm is affiliated with or endorses BdG Advisory.
Where this connects
- How AI changes execution →
Why more execution capacity in fewer hands changes where decision authority has to sit.
- Deciding what should scale →
Where additional people and coordination improve the outcome, and where AI allows a more concentrated human core.
Related perspectives
- The Ratio Is 7:1. Most Enterprises Invert It. →
AI scales only when roughly seven parts of the effort go into governance, controls and authority structures for every one part spent on new execution capability. Enterprises that invert the ratio buy capability they cannot absorb.
- The Operating Model Gap: Why Tool-Led AI Fails →
Organisations approach AI through tools, but AI value depends on structure, incentives and control, so buying the technology without redesigning the operating model reliably produces adoption without results.
- The Intelligent Edge Changes Who Decides →
When inference runs on site, the decision runs on site, and most enterprises have moved their decisions to the edge without moving the decision rights that should govern them.
Begin with the decision
Close the gap between AI ambition and accountable execution.
Bring the situation, the constraints and the decision you are facing.
The other engagements