You are about to commission an AI engagement, or the commission has already been signed. The proposal describes a path toward an AI operating model. That phrase appears in the executive summary, in the stated objectives, in the success criteria. It appears because it names what the organisation needs.
Survey the fifteen largest AI practices, from McKinsey to mid-market specialists, and "operating model" features in strategic communications, white papers and CEO statements from every one of them. What those practices formally name their programmes is a different list.
McKinsey's QuantumBlack is a platform of 25 development tools and 300 research and development accelerators. Accenture's AI Refinery converts raw AI capability into enterprise-grade operational systems. IBM's Garage Method is a prescriptive delivery methodology combining Design Thinking, Lean and Agile practices. Deloitte's Enterprise AI Navigator produces a roadmap; its AI Academy builds workforce capability as a separate service line. Each is precise about what it delivers.
The operating model framing in the proposal describes an ambition. The formal programme scope describes what the engagement is designed to produce. The gap between the two is where the design question lives.
Sources: McKinsey QuantumBlack (mckinsey.com); Accenture AI Refinery (accenture.com); IBM Garage (ibm.com); Deloitte AI Academy (deloitte.com); BCG AI at Scale (bcg.com); Slalom AI Office Blueprint (slalom.com); and nine additional practices surveyed.
The firms' own published research is consistent on the key questions. BCG's work on AI at scale attributes 70 per cent of the value produced by AI investment to new behaviours and working methods, rather than to technology. The implication is precise: the majority of the return is governed by how the organisation is designed to use AI, sustained by the operating structures built around it.
Source: Scaling AI Requires New Processes, BCG 2026
EY built its wavespace network of 45 immersive delivery centres after concluding that political and emotional barriers are the primary obstacle to AI transformation. The wavespace creates the conditions in which leadership teams make the commitment in practice. That diagnosis names the constraint with precision: the limiting factor is leadership behaviour, governance commitment and organisational will. It is an operating model argument, delivered through an experience design product.
Source: EY wavespace
Deloitte offers its AI Navigator roadmap and AI Academy workforce programme as two distinct service lines. The organisations that compound from AI investment are those that integrate both within a single design effort: operating model design and workforce capability built together, around the same value streams, in the same programme. The market has structured these as separate commissions. The evidence points toward a single design problem.
Source: Deloitte AI Academy; State of AI in the Enterprise, Deloitte 2026
BCG's research attributes 70 per cent of AI transformation value to new behaviours and redesigned operating structures. The majority of the return is governed by how the organisation is designed to use AI: by the architecture built to sustain it.
Thoughtworks represents the closest framing to an operating model argument in the boutique tier. Their AI/works programme uses compounding language explicitly: "organisational muscle" built through a thin-slice implementation model, where capability accumulates incrementally and is retained structurally across the organisation. The engineering transformation base is a strong foundation for technology-led organisations; operating model design for the broader leadership team is the natural extension.
Source: Thoughtworks AI-First Software Engineering Transformation
The pattern across all fifteen practices is consistent. The research identifies the operating model as the organising architecture: the structure that connects platform, governance and workforce capability into a system that compounds. Each practice addresses one face of that architecture with genuine rigour. Platform delivery, methodology, governance design and workforce upskilling are each valuable. The operating model is the architecture that holds them in relationship and produces compounding value over time.
Before any AI engagement is scoped, one question is worth asking clearly. At programme close, what will the organisation own?
Platform access is an answer. A trained workforce is an answer. A governance framework is an answer. A roadmap is an answer. Each is legitimate and valuable. An operating model is the architecture that connects and compounds all of them.
When workforce capability is built inside a redesigned workflow, under a governance framework the organisation owns, the result compounds. The Prompt Library built in the first session is still in use in month twelve. The working prototype deployed in week ten becomes the foundation for the next workflow redesigned the following quarter. The Governance Framework signed off by the CEO governs the organisation's AI activity after the programme ends. Each deliverable compounds the next.
A strong answer to the ownership question is specific: 28 named deliverables that describe the organisation's AI-enabled operating model, built by the organisation's own leadership team during the programme, owned and compounded by the organisation independently after it concludes. The 28 deliverables span seven categories.
They include two working prototypes in live operations, a Governance Framework the CEO signs off on, a Prompt Library built and tested across 12 sessions and a Scaling Roadmap the organisation runs independently. Every deliverable is produced by the organisation's own people during the programme. Every deliverable stays.
The question is worth asking before the contract is signed, because the design choices made at scoping determine what the organisation holds at the end of 12 weeks. Platform access is what an external party provides. An operating model is what an organisation builds. Both are real outcomes. The question worth asking first is which one the engagement is built to produce.