McKinsey & Company
McKinsey’s “Where AI Will Create Value and Where It Won’t” makes one core argument worth pulling forward: companies misread AI’s strategic value because they default to the productivity frame. Faster spec writing, faster code, faster proposals. These are floor effects. The ceiling moves only when AI changes what a team can offer or how the business is organized. Most leaders never move past floor effects.
The article frames three waves: productivity gains, differentiation through new offerings, business-model reinvention. The hierarchy is right.
Productivity gains lift the floor. The cost structure underneath the work moves. The ceiling does not move unless the compression is reinvested. Most companies harvest the savings as ROI and stop. This rarely expands profit pools.
Productivity becomes offering expansion when the compressed cost lets a team take on work it could not previously serve. In a model-based engineering workflow, AI inside L100-L300 development lets a smaller team carry a larger program. The team’s addressable scope expands: more requirements complexity, more architecture variants, more verification. The program ladder moves up a rung. That is no longer productivity in McKinsey’s frame; that is differentiation, riding on internal AI adoption that looked like productivity in the budget. Most companies cross this bridge without noticing.
Business-model reinvention is where McKinsey is least specific and where the actual money is. Three concrete shapes:
From product to outcome. Industrial customers pay for uptime, throughput, or yield instead of hardware-plus-services. AI makes outcome SLAs underwritable. From subscription to consumption. AI-mediated services scale with use; unit economics dominated by inference-cost trajectories, not seat counts. From license to embed. Software that sold as a product becomes a callable model inside someone else’s workflow. The original vendor either owns the embedding API or loses the surface.
Each is a different pool than the company is in today. McKinsey calls this reinvention. Survivors move to a new pool; they don’t optimize the old one.
Two core implications.
First, internal AI adoption is the cheapest experiment for testing whether your team can expand offerings. Before reinventing the business model, prove you can take on adjacent program scope. If you cannot, the reinvention thesis is theoretical.
Second, the transaction-cost lens (which should have been the article’s own thesis) is the real warning. Agentic systems collapse information asymmetry. Value migrates to control points: customer interface, data ownership, distribution. Incumbents who see this and position will look back on 2025-2027 as the window. Companies still treating AI as faster-spreadsheets will find their pool inside someone else’s by then.