A new global study offers a useful correction to the easy story that artificial intelligence either “works” or does not. The evidence suggests that many organizations can prove value in a contained project. Far fewer can turn that proof into an operating model that matches the original business case.
BearingPoint’s study, released October 1, surveyed 1,050 C-suite executives and senior leaders across 13 countries in Europe, the United States and China in August 2026. Among organizations that had implemented AI, 74% reported a measurable revenue or cost impact. Yet only 13% said they had scaled their initiatives completely in line with the original business case.
Those are self-reported survey results, not audited returns. But the contrast is still decision-useful: a successful pilot is evidence that a use case may work. It is not proof that enterprise-scale economics, controls or workforce changes will follow.
The judgment
Implication: The finance question is shifting from whether an AI pilot produced value to whether the organization can reproduce that value after integration, controls, adoption and operating-model costs are included.
What would change the conclusion: Verified results showing that scaled deployments sustain adoption, deliver recurring benefits net of full run-rate costs and convert freed capacity into measurable output, service or expense improvement would make the scaling gap less concerning.
Management action: Require a separate scale decision after the pilot. The approval memo should restate the baseline, include full implementation and operating costs, identify the process owner, explain control changes and specify how capacity created by AI will become an economic result.
Positive returns can coexist with failed scale
The study found more evidence of cost reduction than revenue creation. Among 685 respondents with implemented AI, 24% reported cost reductions of at least 10%, while 4% reported revenue or service-delivery gains of at least 10%. BearingPoint also reported that nearly half of organizations said current AI impact was below 4% of costs and below 2% of revenue.
That does not mean small gains are unimportant. A modest improvement in a frequent, expensive process can be worth funding. It does mean that “AI is delivering value” is too broad to support a capital decision. Finance needs to know which workflow improved, how the baseline was measured, whether the benefit recurs and what additional investment is required to expand it.
The distinction matters because the cost curve changes at scale. A pilot can use a limited dataset, a motivated team and manual workarounds. Enterprise deployment adds data cleanup, systems integration, identity and access controls, monitoring, training, process redesign and support. Those costs are often prerequisites rather than failures, but they belong in the business case.
Capacity is not the same as savings
The most consequential finding may be organizational rather than technical. Sixty-two percent of respondents reported AI-induced workforce overcapacity of at least 10% today. That figure should be treated cautiously: it reflects executive estimates, not a verified measure of idle labor or a recommendation to remove positions.
Still, it exposes the missing bridge between productivity and financial value. If a team completes the same work in fewer hours, the organization has created capacity. Value appears only when management decides what happens next. The team might process more volume without adding staff, improve service, shorten cycle times, reduce outside spending, absorb vacancies or redirect people toward higher-value work.
If nothing changes, the capacity is absorbed. Employees remain busy, costs remain in the forecast and the business case records a theoretical benefit that never reaches the income statement or service model.
This is why AI investment needs an operating owner, not only a technology sponsor. Finance can validate the baseline and track the economics, but the accountable leader must redesign the workflow, set adoption expectations and decide how the released capacity will be used.
Scaling deserves its own capital gate
Organizations often approve a pilot and then treat expansion as the next implementation phase. A better approach is to treat scale as a new investment decision.
The scale memo should answer five questions:
- Did the pilot solve the original problem at an acceptable quality level?
- What is the full cost of data, integration, security, controls, training and support?
- Which benefits are already measured, and which remain assumptions?
- Who owns adoption and the conversion of capacity into results?
- What evidence would pause, redesign or stop further funding?
This extends the staged-funding framework in The CFO Is Becoming the AI Capital Allocator. A technically successful pilot may deserve more capital, but only if the economics survive the move from a controlled test to the actual operating environment.
Governance must scale with autonomy
BearingPoint found that most organizations remain early in agentic architecture: 13% reported a defined strategy with active initiatives, and 10% said they were scaling agentic architecture across the organization. As AI systems gain permission to access data, initiate actions or interact with other systems, the risk does not remain at pilot size.
The control design therefore needs to change before autonomy expands. Permissions, data boundaries, logs, exception handling, human review and accountability should be explicit. The practical principle is the same one described in AI Governance in Finance: adoption and governance should arrive together.
Leaders in BearingPoint’s maturity framework were much more likely to connect projects to measurable financial KPIs. Seventy percent of leaders linked more than half of their AI projects to financial measures, compared with 34% of implementers. That association does not prove that measurement caused scaling success. It does suggest that accountability, data readiness and operating discipline are part of the capability—not administrative work added after the technology is chosen.
The board needs a scale view, not a project count
A board dashboard that reports the number of AI projects, licenses purchased or employees trained can make activity look like progress. Decision-useful reporting should instead separate experiments, implemented workflows and scaled operating capabilities.
For material initiatives, show the approved business case, cumulative spending, full run-rate cost, adoption, realized benefit, remaining assumptions, control status and the next capital gate. The free Board Finance Dashboard can be adapted to place those measures beside liquidity, forecast pressure and other warning signals.
The study’s central lesson is not that AI has failed to create value. It is that value found in a pilot is not self-executing. Scaling requires management to connect technology to data, controls, workflow, workforce and capital allocation. That is where the business case becomes real—or quietly disappears.
Sources and notes
- BearingPoint, Scaling AI for measurable impact, September 2026; press release dated October 1, 2026. Survey conducted online in August 2026 among 1,050 C-suite executives and senior leaders.
- Reuters, October 1, 2026, summarizing the study’s scaling, integration and regulatory findings.
Leave a comment