AI Doesn't Create Value. Decisions Do.

In Part 2 we measured the AI Value Realization Gap: 88% of organizations use AI, only ~6% capture material EBIT impact from it. Today, the question that follows: where exactly does the value get lost?

Not in the models. In the chain that comes after them.

The AI Value Chain

AI produces exactly one thing: intelligence potential. Predictions, recommendations, alerts, generated artifacts. That potential becomes value only by surviving four conversions:

Intelligence → Decisions → Actions → Value → Reinvestment

• A prediction nobody uses to decide is a report.
• A decision nobody executes is a memo.
• An action nobody measures is an anecdote.
• Value nobody reinvests is a one-off.

The chain is multiplicative

Conversion efficiency compounds link by link: Φ ≈ DCR × ACR × VCR. Recall the bank from Part 2: 40% × 70% × 50% ≈ 14% of potential reaches the P&L.

Multiplicative has a hard consequence: excellence at three links cannot compensate for a broken fourth. Value flows at the rate of the weakest link.

That inverts the usual investment logic. In that bank, doubling model accuracy moves almost nothing; the bottleneck is upstream of the model's quality. Raising DCR from 40% to 60% (decision rights, ownership, workflow integration) lifts realized value by 50%. Same models. Same data. Different chain.

Why we miss it

Because we industrialized the production of intelligence (platforms, copilots, agents) and left its conversion unmanaged. Chains rarely break loudly; they leak:

• Decision rights nobody redefined when AI arrived.
• Pilots without a P&L owner.
• Recommended actions that compete for resources with the annual plan, and lose.
• Value that is never isolated, measured, or attributed.
• Gains consumed as one-offs instead of reinvested in capabilities.

One upstream note: the chain converts whatever enters it. The Expert Validation Loop in AI-native engineering, validating the knowledge agents produce and not just the code, matters here: converting incorrect intelligence faster is not progress, it's accelerated risk.

What's next

Each link gets an instrument: DCR, ACR, VCR, AIRR, plus one upstream constraint that throttles all of them: ALDI, the AI Leadership Debt Index.

Where does your AI value chain break first, and who owns that link?