Most organizations are not struggling to adopt AI.
They are struggling to classify it correctly.
And once an AI initiative is misclassified, it is governed incorrectly from the start.
THE PROBLEM
In practice, many AI investments are routed through familiar channels: IT budgets, innovation pipelines, digital transformation programs.
These structures are not inherently flawed. They exist for a reason.
The issue is that AI does not consistently behave like the investments those structures were designed to manage.
AI initiatives vary along two dimensions that materially affect how they should be governed:
Reversibility — how easily a decision can be undone
Uncertainty — how predictable the outcomes are
This creates a different type of capital profile than traditional projects.
THE FRAMEWORK
The AI Capital Classification Matrix
AI initiatives can be understood through four categories:
Low Uncertainty / High Reversibility Small pilots, internal tools, contained use cases. → Minimal governance, rapid iteration. |
High Uncertainty / High Reversibility Experimental models, early-stage use cases. → Structured experimentation, controlled risk. |
Low Uncertainty / Low Reversibility Embedded systems, workflow automation, operational integrations. → Strong process alignment, execution discipline. |
High Uncertainty / Low Reversibility Customer-facing AI, decision automation, large-scale deployments. → Highest governance, cross-functional oversight, risk controls. |
This framework is not about increasing governance. Customer-facing AI, decision automation, large-scale deployments. → Highest governance, cross-functional oversight, risk controls. |
This framework is not about increasing governance. It is about matching governance to the actual risk profile of the investment. In lower-risk categories, this often reduces unnecessary oversight. In higher-risk categories, it prevents under-governed exposure. |
WHERE RISK ACCUMULATES
This distinction is not theoretical.
In capital review discussions and integration settings, what tends to happen is this:
An AI initiative is approved under one assumption — typically as a reversible experiment — but evolves into something far less reversible over time.
The governance structure does not evolve with it.
That is where risk accumulates.
Public examples reflect this pattern.
Efforts like IBM Watson Health were not constrained by lack of investment or ambition. They struggled with uncertain outcomes paired with decisions that became increasingly difficult to unwind.
This is not a failure of technology alone. It is a mismatch between how the investment behaved and how it was governed.
MONDAY QUESTIONS Three questions to bring to your next capital review or governance meeting. 1. If this AI investment underperforms, how quickly could we unwind it without operational disruption? 2. Which assumptions about this system are most likely to change over the next 12–24 months? 3. Are we approving this investment based on expected ROI, or on an explicit understanding of its reversibility and uncertainty? |
WHAT’S NEXT
Next week’s episode examines a case most organizations reference but few fully understand.
The failure was not technical. It was a governance decision that went unchallenged.
AI is a capital allocation and governance decision that happens to involve technology. |
Strategic Risk Lab
AI Strategy · Risk · Governance · Capital Allocation