AI Is Improving Productivity

AI Is Improving Productivity. Why Isn't More of It Showing Up in the P&L?

The next challenge for enterprise AI is not proving that employees can work faster. It is showing what the business does with those gains.

Enterprise AI is getting better at producing one kind of evidence: productivity.

Employees can draft reports faster. Customer service teams can handle more work. Developers can shorten parts of the software development cycle. AI usage is spreading across organizations.

But the question changes when it reaches the CFO:

McKinsey's 2026 global AI survey found that 80% of respondents said AI had improved their individual productivity. Yet only 37% said AI had contributed at least some positive impact to their organization's EBIT.

EY Global AI Leader Dan Diasio recently described a similar gap at The Information's AI Agenda Live. He said that, among the companies he sees, only about one in ten can point to where the return from AI is showing up in the P&L.

That figure does not mean 90% of companies have no AI ROI. It was Diasio's observation, not the result of an EY global survey.

But it sharpens the question:

Saving time is not the same as capturing value

Consider an employee who used to spend 60 minutes completing a task and can now finish it in 40 minutes with generative AI.

That is evidence of a productivity improvement.

The remaining 20 minutes, however, are not automatically revenue or lower costs. Their economic value depends on what happens next.

Does the employee use that capacity to handle more work? Does the company reduce external spending? Can the business grow without adding staff at the same rate? Does the faster process shorten the path to revenue?

Or does the organization continue operating much as it did before?

This is why several measures commonly grouped under “AI ROI” need to be separated.

Adoption metrics tell you whether people are using AI. Productivity metrics tell you whether work is getting faster, better, or less labor-intensive.

Neither, on its own, establishes that the organization has captured an economic return.

The missing question is operational:

The gap may be in how the business operates

McKinsey's survey provides an important clue.

Only about 6% of respondents qualified as AI high performers, defined in part by attributing at least 5% of EBIT impact to AI while also reporting significant value from the technology.

These organizations were more likely than others to fundamentally redesign workflows around AI rather than simply insert AI tools into existing ways of working.

That does not prove that workflow redesign causes higher AI returns. The evidence shows an association, not a universal causal relationship.

But it gives executives a useful way to examine their own deployments.

Suppose a Copilot rollout saves employees an average of three hours a week.

The next question should not simply be:

It should be:

If customer service employees use that capacity to handle more cases, the company can examine service capacity and cost per case.

If sales teams use it to engage more prospects, the business can track whether the additional activity affects pipeline or revenue.

If the company can support growth without adding staff at the same rate, it can examine avoided hiring and changes in unit economics.

If none of the underlying operating model changes, the productivity improvement may still improve employee experience or create useful capacity. But that is different from demonstrating a financial return.

“Positive ROI” can mean very different things

EY's own research illustrates how easily the language of AI ROI can become ambiguous.

In its July 2026 US AI Pulse Survey, 98% of senior leaders whose organizations were investing in AI said they were seeing positive ROI.

At first glance, that appears difficult to reconcile with Diasio's observation that only about one in ten companies he sees can point to where AI returns appear in the P&L.

The two findings do not necessarily conflict.

Reporting a positive return from an AI investment and identifying its financial impact in the P&L are different standards of evidence.

So when an internal report says an AI project has achieved “positive ROI,” executives may need to ask one more question:

Does it mean perceived value, calculated time savings, higher productivity, an operational improvement, or a measurable change in revenue, cost, or margin?

Without that distinction, organizations can use the same term to describe very different levels of evidence.

Productivity does not have to become headcount reduction

There is another reason not to reduce AI ROI to cost cutting.

EY's late-2025 AI Pulse Survey found that 96% of companies investing in AI reported productivity gains. But organizations were not necessarily converting those gains directly into workforce reductions.

Some were reinvesting the capacity in AI capabilities, R&D, cybersecurity, and employee retraining.

That matters because productivity can create value in several ways.

A company can use it to lower costs. It can also use it to increase service capacity, accelerate product development, support more customers, or redirect resources toward growth.

The management question is therefore not:

It is:

That is the trail executives need to follow if they want to understand where AI value is actually being created.

ROI also has a cost side

Productivity metrics can obscure another part of the equation.

Saving three employee hours a week does not establish positive ROI simply by multiplying those hours by salary.

Enterprise AI carries costs beyond model usage. EY's recent analysis of agentic AI ROI points to infrastructure, software, governance, organizational change, expected failures, and regulatory requirements as costs that may need to be considered alongside model or token spending.

So enterprises need to trace both sides of the equation:

• What economic effect did the productivity gain create?

• What did it cost to produce that effect?

Without both, an ROI calculation can overstate the business value of the deployment.

The better question: Where is the value now?

None of this means every AI pilot should immediately be judged by its impact on the P&L.

During experimentation, measures such as completion time, quality, error rates, or output can be entirely appropriate. They answer an important first question:

The standard should change as the deployment matures.

Once a company begins expanding licenses, rolling AI across departments, increasing budgets, or embedding the technology into regular operations, productivity alone becomes less sufficient as evidence of business value.

Executives then need to distinguish among three questions:

1. Did the work improve?

2. What changed operationally because of that improvement?

3. What economic effect did that operational change produce?

If a company can answer only the first, it may have demonstrated that AI is useful.

If it can answer the second, AI is beginning to change how the business operates.

The third is where the organization begins to build a stronger case for enterprise-level financial return.

For companies moving from pilots to scaled deployment, the next stage of AI value measurement is not another adoption metric or another estimate of hours saved.

It is being able to trace what happened after those hours were saved.

Decision Judgment

Productivity improvement can be a reasonable success metric during experimentation and pilot programs.

Once AI moves into scaled investment and regular operations, enterprises need additional evidence showing how those gains change workflows, resource allocation, service capacity, costs, or revenue.

Turning Global AI Signals into Business Decisions

InfoAI helps decision-makers identify what is changing in AI, evaluate the evidence behind those changes, understand what they mean, and determine which business assumptions or decisions may need to be reconsidered.

www.InfoAI.com.tw