Deploying AI creates no advantage.
How you decide
creates one.
tointelligence · omer taki
· the AI value paradox

AI creates value, but not necessarily for you.

AI undeniably creates economic value. But this value is distributed very asymmetrically. Major model vendors capture a disproportionate share. Integrators capture the second share. Deploying organisations, often, the third.

Why? Because AI value comes primarily from the models themselves, and these are proprietary. When an organisation uses GPT-4 to gain productivity, it pays OpenAI to access that value. It does not create it; it purchases it.

Buying AI productivity is not the same as building an AI competitive advantage.
· three deployment levels

Not all AI deployments create the same value.

level 1
Generic automation
Standard AI tools for repetitive tasks. Real but temporary productivity gains : competitors do the same. No defensible advantage.
level 2
Differentiating integration
AI integrated with proprietary processes or data. Insights difficult to replicate. Advantage temporarily defensible but requires active dependency management.
level 3
Proprietary capability
Internal AI capabilities built on proprietary data. Advantage is structurally defensible because competitors cannot replicate without the same data.

Most organisations are at level 1. They create value, but it is immediately accessible to their competitors at the same price. It is a commodity investment, not a competitive advantage.

· where value is destroyed

AI also destroys value : often invisibly.

· destruction via hidden costs

Licences rising after lock-in, underestimated integration costs, team training and adaptation, EU AI Act governance and compliance. These transform announced positive ROI into real negative or marginal ROI. We observe that AI governance costs are systematically underestimated by a factor of 2 to 3.

· destruction via dependencies

Every AI integration creates dependencies. These have a future cost: exit cost, lost options, exposure to changing vendor conditions. This future cost must be integrated into current value calculations. It almost never is.

· conditions for real value

When AI creates defensible value.

AI value is defensible when it rests on three simultaneous conditions: proprietary data competitors don't have, augmented decision capability inaccessible without that data, and dependencies managed so they don't become constraints.

The difference between an organisation that captures AI value and one that finances it for its vendors is the quality of its strategic decisions.