AI concepts
AI Operating Model (a company's operating model for AI)
The way an organisation decides, builds, deploys, and maintains AI solutions at scale. Four main models: a central AI unit (Center of Excellence), a federated distributed model, a hybrid model, and AI embedded in business units. The choice of model determines 60 to 80 percent of the long-term ROI from AI in a company.
Primary source: McKinsey Operating Model for AI 2025, BCG Build vs Buy AI 2024, Deloitte AI Institute 2025
In 2023 and 2024 the dominant model was experimental, meaning every department tested AI independently. In 2025, firms with measurable AI ROI already had a formalised operating model. Firms without a formalised model generated the operational cost of duplicated deployments (the same tools bought by three departments), governance inconsistency, and suboptimal use of vendor contracts.
Four basic models
McKinsey Operating Model for AI, July 2025, an analysis of 320 enterprise firms, describes four archetypes:
Centralized AI (Center of Excellence). One unit responsible for all AI deployments in the company. Pros: consistency, scale, governance. Cons: bottleneck, distance from the business. Used in firms with strong central IT, populations of 1,000 to 5,000 people.
Federated AI. Each business unit builds its own AI with minimal central support. Pros: speed, closeness to the business problem. Cons: duplication, lack of standardisation, security and compliance risks. Used in firms with highly diversified units.
Hub-and-spoke. A central AI team sets standards and the platform, business units build their own applications on that platform. Pros: a balance of scale and speed. Cons: requires a mature organisation. The most frequently recommended model for firms of 500 to 3,000 people.
Embedded AI. AI specialists distributed across business units, with no central unit. Pros: maximum closeness to the problem. Cons: no governance, no scale. Used in startups or very small firms (under 200 people).
The selection decision
BCG Build vs Buy AI from 2024 points to three variables. First, the firm's maturity in IT management (low = centralized, high = federated or hub-spoke). Second, business diversification (homogeneous = centralized, diversified = hub-spoke). Third, the availability of AI talent (low = centralized, high = federated).
Polish context
Deloitte AI Institute Poland 2025 estimates that among Polish firms of 500-plus people, around 65 percent operate in the experimental model (that is, no model), 20 percent in centralized, 10 percent in federated, and 5 percent in hub-and-spoke. Most firms are two to three years behind the global median of operating-model maturity.
Choosing and implementing an AI Operating Model for your company is part of the AI Readiness Audit plus, optionally, internal IT support.