Companies across Asia-Pacific are ramping up artificial intelligence investments at a pace that far exceeds their ability to measure the returns, according to a new KPMG survey. The gap between spending and proof is becoming a critical challenge for CIOs and CFOs as budgets balloon and boards demand accountability.

KPMG surveyed more than 2,100 senior executives globally, including 521 respondents from Australia, China, India, Japan, South Korea, and Singapore. The findings paint a picture of an AI market driven by enthusiasm and competitive pressure, yet still immature in its measurement practices. The divide between those who see value and those who can prove it is stark.

AI Budgets Surge But Proof Lags

The survey found that 70% of companies in the Asia-Pacific region plan to invest more than US$50 million in AI over the next 12 months. At the same time, only 5% said they have established return on investment with demonstrated business outcomes. That single statistic captures the core tension of enterprise AI today: money is flowing in faster than evidence is coming out.

Despite the low proportion with formal ROI, 81% of APAC companies surveyed said AI is already delivering meaningful business value through productivity gains, cost savings, or revenue growth. That figure rose from 69% just three months earlier, indicating that perceived benefits are spreading quickly. Yet perceived benefits are not the same as measurable returns. Executives may see improvements in their daily operations, but finance teams need numbers that can be verified and defended in budget reviews.

KPMG's data suggests a measurement problem rather than a value problem. Many organizations are confident that AI helps, but they lack the mechanisms to quantify how much. This gap becomes harder to ignore as investments grow and as cost pressures mount from more complex AI systems.

Adoption Remains Shallow in Many Markets

The unevenness of AI maturity across the region is striking. India reported the highest level of business value from AI at 89%, followed by Australia at 86%. But these self-reported figures often mask the reality of adoption depth. Singapore's own data, published by the Ministry of Manpower in April, shows how early the process still is for many firms.

MOM reported that 71.5% of private-sector establishments with at least 10 employees had not adopted AI at all. Only 3.8% had integrated AI into their core business processes. That suggests that most of the value being reported comes from limited, experimental use cases rather than enterprise-wide transformation.

Adoption is heavily skewed toward larger companies. MOM found that AI adoption rose from 23.9% among firms with fewer than 25 employees to 76.4% among the largest firms. Smaller firms face higher barriers including cost, talent, and data readiness. Larger firms have more resources to experiment, but they also face more complexity in measuring what works and what does not.

For companies already using AI, productivity remains the clearest reported gain. MOM said 70.7% of AI-adopting Singapore firms reported improved worker productivity, while fewer cited better decision-making or innovation. This pattern is common elsewhere in APAC: AI is easiest to apply to routine tasks, where output can be compared with a baseline.

Cost Visibility Is Uneven

Tracking AI costs is another area where the region is split. KPMG found that about 80% of APAC respondents have full or partial visibility into AI operating costs. But that visibility is not the same as real-time control. Australia is further ahead, with 40% of companies monitoring AI costs fully and in real time, compared with just 12% in South Korea.

The gap between awareness and control matters because AI costs are not static. Token consumption, compute usage, and human review time can all vary dramatically depending on how systems are used. Without a clear view of these costs, companies cannot predict budgets or identify inefficiencies.

The cost pressure is already forcing difficult decisions. KPMG said 55% of APAC companies had delayed or scaled back AI agent rollouts because operating costs began to exceed the value generated. This trend is significant as companies move from simple chatbots to more expensive agentic systems that can take actions autonomously. These systems consume far more tokens and often require additional human oversight, making cost control more urgent.

Many companies are still investing in foundations rather than advanced use cases. KPMG said APAC companies are putting AI budgets toward IT infrastructure, cyber and data security, operations, and transformation. These investments may be necessary, but they do not always create quick, measurable ROI. Infrastructure spending, in particular, is difficult to tie to a specific business outcome because it supports multiple initiatives at once.

Bringing Discipline to AI Measurement

Consultancies are increasingly arguing that AI cost management requires a new level of granularity. Accenture's tokenomics report defines the discipline as connecting what AI consumes to the value it returns. The firm found that in nearly every deployment it examined, fewer than 10% of users and workflows drove most of the AI bill. This concentration of usage means that small changes in behavior can have a significant impact on cost.

Accenture also discovered that if token prices fell by 25%, only 15% of organizations would actually bank the savings. Most would reinvest the difference into more AI use. This suggests that AI spending is generative: it expands to fill available budget unless companies set clear boundaries and continuously measure against expected returns.

BCG makes a similar point in its work on return on AI, or RoAI. The firm advises companies to track AI costs at the workflow level, including token use and human review time, then compare that cost with the business outcome produced. This approach shifts attention away from aggregate metrics and toward the specific processes where AI is being applied.

For IT and finance leaders, the first move does not need to be complicated. Pick one expensive or high-value AI workflow, record the pre-AI baseline, and track token costs, human review time, quality, and business output over a set period. The review should end with a clear decision: scale it, redesign it, or stop it. This simple discipline creates a repeatable structure for evaluating AI investments across the organization.

The need for such rigor is growing as AI adoption expands beyond experimentation. OpenAI's business adoption and enterprise AI use continue to rise, and regulators are paying closer attention to how autonomous systems operate. Australia, for example, is developing specific AI rules that could require companies to demonstrate that their systems are reliable and accountable. Without solid ROI data, compliance becomes harder and budget decisions become more contentious.

KPMG's findings do not mean AI is failing in APAC. They show that enthusiasm, adoption, and reported productivity gains are ahead of formal measurement. Companies that close that measurement gap will have a clearer view of which AI projects deserve more budget and which ones should be cut before costs keep rising. The competitive advantage in the next phase of AI will not belong to the biggest spenders but to those who can prove what their spending actually returns.

Source: TechRepublic News