112% Higher AI ROI — Yet 77% Say AI Adoption Is Outpacing Governance

IBM recently published two numbers that are difficult to put next to each other.

Organizations that use AI performance data to guide funding and talent decisions report 112% higher AI ROI than their peers. Yet only 25% of tech executives consistently measure AI ROI.

That is already a remarkable gap. But then the numbers get more uncomfortable:

• 85% of technology leaders lack real-time visibility into AI spending.
• Cloud costs are running 48% above initial projections.
• 77% of executives say AI adoption is moving faster than their governance capabilities, with IBM warning that AI can succeed technically while still failing commercially.

Put those findings together and a pattern starts to emerge. Companies are getting better at deploying AI. They are not necessarily getting equally good at deciding:

• What is working?
• What is it costing?
• What deserves more capital?
• What needs to change?
• What should stop?

Measurement, cost visibility, and governance can look like three separate AI problems. I think they may be three symptoms of the same one:

Organizations are deploying AI faster than they are building the decision systems needed to turn evidence into action.

And that changes what the AI ROI problem really is. The question is no longer simply: Can we measure the return?

It is:

Can we use what we learn to make the next investment decision better?

Measurement should change the next decision

ROI is often treated as something that gets calculated after the work is done. A project launches. The team reports results. Finance reviews the numbers. Someone decides whether it was successful.

But IBM’s findings point toward a much more useful role for measurement. Performance data should influence funding and talent decisions.

That changes the operating loop:

Define value → measure performance → evaluate evidence → decide → reallocate → measure again

The important part is not the ROI number itself. It is what the number changes. A strong measurement system should help leaders answer:

• What deserves more funding?
• What needs redesign?
• What should be constrained?
• What should stop?
• Where should talent move?

IBM also reports that 68% of Chief AI Officers launch initiatives even when they lack a reliable way to assess impact. That is more than a reporting weakness. It means capital can remain attached to weak initiatives simply because nobody has built a disciplined mechanism for challenging them.

The value of measurement is not the score. It is the decision the score makes possible.

Cost visibility should change where capital goes

Value is only half the equation. The other half is what it costs to produce that value.

AI economics move quickly: usage changes, cloud demand shifts, new models appear, prices fall, and workloads grow. A system that looked economically attractive six months ago may no longer be the best architecture today.

That is why IBM’s 85% cost-visibility figure matters. If leaders cannot see what AI is actually costing, they cannot confidently answer:

Which initiative deserves the next dollar?

A premium model may be justified for a complex, high-consequence workflow. It may be wasteful for a repetitive task that a smaller model can handle just as well. IBM argues that organizations should treat AI as a living portfolio, continuously comparing spending, ownership, performance, and expected returns.

That turns cost management into a decision loop:

measure cost → compare outcome → evaluate alternatives → reallocate → measure again

• If a cheaper model produces the same outcome, route the work there.
• If costs rise faster than value, intervene.
• If the economics improve, scale selectively.
• If they deteriorate, redesign or constrain.

Visibility matters because it changes allocation.

Governance should determine authority

The third problem is governance. But governance here means much more than compliance.

IBM argues that organizations need explicit decision rights, intervention mechanisms, audit trails, escalation rules, accountability, and human judgment embedded into the operating model. That means answering questions like:

• What can AI recommend?
• What can it execute?
• When must a human intervene?
• Who owns the outcome?
• What happens when confidence falls?
• When should authority expand?
• When should it contract?

As AI moves from generating information to taking action, those questions become operational. The more capable the system becomes, the clearer the authority model has to be.

And governance also affects economics. IBM explicitly connects governance to model selection and periodic reassessment of whether the current architecture still makes sense. So governance affects:

• Risk
• Cost
• ROI
• Capital allocation
• Autonomy

That is why these are not really separate problems.

Measurement, cost, and governance converge on the same decision

This is the part I think matters most: measurement without action is just reporting, cost visibility without allocation is just accounting, and governance without decision rights is just policy. The value appears when the three connect.

A decision system should be able to answer:

• What evidence do we have?
• How strong is it?
• What is this initiative costing?
• What value is it creating?
• What risks matter?
• Who has authority?
• What action is justified?
• What should happen next?
• What outcome should cause us to revisit the decision?

That produces a much more useful operating loop:

evidence → judgment → decision → action → outcome → revision

Then repeat. Because AI will not stand still: models will improve, costs will move, workflows will change, new risks will appear, and better alternatives will emerge. The decision system has to keep up.

This is the operating problem Decision Systems AI is built around

At Decision Systems AI, this is the layer I have been focused on. Not simply "What can AI do?", but "What should the business do because of what AI can do?"

That means connecting evidence, ROI, cost, confidence, risk, governance, authority, action, outcomes, and learning into one decision process.

The goal is not to produce another dashboard. It is to help the organization decide:

• What should scale
• What needs redesign
• What should be routed differently
• What requires tighter governance
• What should be constrained
• What should stop

The AI ROI problem may be bigger than ROI

IBM’s findings suggest that the challenge is no longer simply getting AI into the enterprise. It is deciding what deserves to keep moving once AI is already there.

That requires knowing what is working, what it costs, what evidence supports it, what risks remain, who has authority, and what should happen next.

So the AI ROI problem may not ultimately be a measurement problem. Or a cost problem. Or a governance problem.

It may be a decision-system problem.