The AI Operating Model

You have accepted the hard truth from the previous lesson: roughly 70% of AI value is people and process, not algorithms or data plumbing (BCG 10-20-70). The natural next question is organizational, and it is one only a senior leader can answer:

Where do AI authority, talent, and standard-setting live in your company?

This is not a technology choice; it is an operating-model choice, and it determines whether your AI effort scales or fragments. Get it wrong in one direction and a central team becomes a bottleneck that the business routes around. Get it wrong in the other and every business unit reinvents the wheel, with inconsistent standards and uncontrolled risk.

There are three broad answers, and the evidence increasingly points to one of them. The rest of this lesson is about choosing deliberately — and about the single most-overlooked lever that decides whether the chosen model actually works: your middle managers.

Every AI operating model is a variation on three archetypes. The trade-off is always the same tension: central control versus business-unit speed.

The centralized model is safe but slow; the federated model is fast but ungoverned. Hub-and-spoke — sometimes called a centralized-federated model or a Center of Excellence — is the deliberate middle path, and the 2025-26 evidence says it is what separates firms that scale from those that stall.

Two time-stamped data points (re-verify before quoting — these refresh):

• Firms that successfully scale AI are reportedly ~3x more likely to use hub-and-spoke than any other structure (Dataiku research, cited 2025).
• Chief AI Officers operating in centralized or hub-and-spoke models reportedly achieve ~36% higher ROI than those in fully decentralized structures (IBM research, 2025).

The principle beneath the numbers is durable even as the figures move: federate execution, centralize the guardrails.

The whole model rests on a clean division of labor. Blur it and you get either a bottleneck or chaos. The rule of thumb: the hub owns the 'how-safely' and the 'on-what'; the spokes own the 'what' and the 'so-what.'

Notice what is not on the hub's list: building every use case. A hub that builds everything is just a centralized team wearing a new name — and it recreates the bottleneck you were trying to escape. The hub's product is leverage: platforms, patterns, and permission that make the spokes faster and safer.

The spokes own outcomes because they own the business. A marketing team knows which campaign workflow is worth redesigning; the central hub does not. By making the spoke accountable for the number — revenue, cycle time, CSAT — you ensure AI is aimed at a real business problem rather than a demo.

The most common way a hub-and-spoke model fails is not collapse — it is slow drift from enabler to gatekeeper. It happens for understandable reasons. The central team is accountable for risk, so it adds an approval step. Then another. Soon every spoke project waits in the hub's queue, the business stops feeling ownership, and frustrated teams quietly route around the whole thing with unsanctioned tools.

Watch for these warning signs that your hub has drifted:

• Business units describe the AI team as 'the people you have to get past,' not 'the people who help.'
• The hub's backlog is full of build requests rather than platform and standards work.
• Adoption is flat outside the hub's own projects.
• Shadow AI is rising — a reliable signal that the sanctioned path is too slow.

The fix is structural, not cultural exhortation. Fund the hub to build paved roads (pre-approved tools, templates, guardrails that let teams move fast within policy) rather than to operate tollbooths. Reserve mandatory hub review for genuinely high-risk systems, and let low-risk use cases move on a fast, self-serve track. Risk-based tiering — a chatbot is not a loan-approval model — is what keeps governance from becoming a brake on everything.

Here is the move that quietly decides success — and the one most executives miss. You cannot scale adoption from the center. A brilliant central AI lab can build platforms and prove use cases, but the people who actually change how work gets done are line and middle managers. They set their team's priorities, model what 'good' looks like, and decide whether a new tool gets used or quietly ignored.

The evidence on this is pointed. HBR (2026) documents an executive-versus-manager reality gap: senior leaders see AI's strategic advantage from the top, while middle managers confront its flaws — unreliable output, unclear workflows, no support — largely unsupported. That gap stalls adoption and costs real money. Top-down mandates ('everyone will use AI') make it worse, because they add pressure without removing friction.

The fix is to treat managers as partners in the operating model, not recipients of a rollout:

• Involve managers in planning, so use cases reflect the real friction in their workflows.
• Reduce their admin burden so adoption is easier, not one more thing on the pile.
• Build feedback channels that carry the front line's reality back up to the hub.
• Equip them to lead the change locally, rather than waiting for the central lab to push it.

This is why the spoke owns local change management in the table above. The hub enables; the manager adopts.

An operating model needs a focusing discipline, or the spokes will scatter into a hundred shallow experiments — the exact pattern that produces no measurable impact. The most useful framing here is McKinsey's four-pillar approach for agentic AI value:

1. Pick 2-3 high-value vertical use cases — concentrate, do not sprinkle. Winners go deep on a few; laggards run many pilots that never move a number.
2. Embed AI into reimagined workflows — do not bolt agents onto legacy processes; redesign the end-to-end workflow around them.
3. Scale with infrastructure, cross-functional teams, and strong governance — this is precisely what the hub provides to the spokes.
4. Build a new skills mix — domain expertise paired with AI fluency, so the spokes can run and evaluate their own use cases.

The through-line connecting this to the operating model: the hub provides pillar 3 (the platform and governance); the spokes execute pillars 1, 2, and 4 in their own domain. Each chosen use case should tie to a specific operational number — a cycle time, a conversion rate, a cost line — so the spoke can prove value, not just activity.

Above all of it sits the differentiator no structure can substitute for: executive sponsorship. BCG reports that nearly 100% of 'future-built' firms have deeply engaged C-suites with a unified vision, real funding, and leaders who personally use AI daily; McKinsey finds CEO oversight of AI governance the factor most strongly correlated with bottom-line impact (both 2025; re-verify live). The operating model is the machine — sponsorship is the power that runs it.