Businesses struggling to generate actual returns on investments (ROI) from their artificial intelligence (AI) deployments are making some common mistakes and overlooking the need to establish new ways to measure returns.
In particular, traditional metrics, such as cost savings, productivity improvements, and revenue growth, still are important, but are no longer adequate on their own.
Measures now have to include other key business improvements, said James Wilson, partner in technology consulting and advisory at KPMG in Singapore.
These should encompass quality of decisions made, speed to execution, workflow automation, customer and employee experience, organisational agility, and the ability to scale expertise across the organisation, Wilson said in an email interview with FutureCIO.
Agentic AI enables systems to orchestrate work, make decisions, coordinate activities, and execute tasks with increasing levels of autonomy.
As such, organisations need a broader definition of value that extends beyond labour replacement or efficiency gains, he noted.
Classic financial metrics should be reviewed in assessing AI gains, agreed Dhruv Dhumatkar, NetApp’s Asia-Pacific Japan CTO and head of solutions engineering.
He pointed to intangible returns that may be harder to measure, such as customer satisfaction and retention.
It is difficult to apply financial-specific lenses to evaluate such returns, even though they are just as valuable, Dhumatkar told FutureCIO in a video chat.
It underscores the need to look at the business case, not just the financials, he added, noting that this has prompted some companies to hire new roles, such as chief AI officers, to align the technology to the organisation’s business growth.
These executives also are tasked to bring together cross-functions teams to ensure AI ROI and the necessary metrics are not defined in isolation, he said.
In addition, businesses have to asses reliability, governance effectiveness, and risk-adjusted outcomes, particularly as AI systems become more autonomous, Wilson said.
“The conversation is evolving from how much work AI automates, to how much business value AI helps to create, accelerate, and scale safely,” he said.
Indeed, it will be increasingly critical that companies have a firm grasp on their AI investments.
By 2029, Gartner predicts that 60% of organisations deploying AI will have a dedicated function responsible for mapping AI total cost to value or profit.
The research firm noted that as more turn to agentic AI, managing token consumption will become a business priority as costs spike, making it more difficult to link spending to business value.
“Organisations need greater visibility into AI consumption patterns and stronger governance to balance cost efficiency with value creation,” Gartner said.
It recommended that enterprises tie token usage directly to business value metrics, while implementing clear governance, quotas, and monitoring controls to ensure AI investments deliver measurable returns.
Moving from pilots to real returns
A lack of clear understanding and definition of what they want AI to do, often is the reason companies fail to realise ROI from their AI projects, said Dhumatkar.
These businesses also do not have clearly defined metrics, he added.
Organisations have to start with the business problem they are trying to solve, not with the technology itself, Wilson concurred.
He added that a common challenge enterprises face is expanding a successful AI pilot to achieve enterprise-wide value.
“Scaling AI requires changes to processes, ways of working, governance, and workforce behaviours,” he noted. “If adoption is low or business processes remain unchanged, even technically successful AI solutions may fail to generate meaningful returns.”
“Ultimately, AI creates value when it is embedded into how the organisation operates, not when it sits alongside existing ways of working,” he said.
In addition, AI models are only as good as the data that feeds them, said Dhumatkar.
If the information is not prepared or tagged correctly, it will not deliver any true merit however good the intent is, he added.
This also highlights the need to make good infrastructure choices, spanning the compute and network systems needed to facilitate smooth data flow to power AI workflows, he said.
These include having good use cases and models that are optimised for the use cases, said Elaine Chan, NetApp’s Asia-Pacific director of AI solution sales.
She noted that ROI comes from different aspects, including concrete financial returns and intangible benefits such as productivity gains.
In healthcare, for instance, ROI can be defined as improved patient outcome and better use of doctors’ time to focus on more beneficial outcomes.
Across the different benchmarks, data remains the core component and must be ready to drive the processes, Chan told FutureCIO.
“You may have optimised compute [resources], but don’t have the right data, and [can end up] not having the right outcome,” she said.
Companies fail to extract ROI from their AI initiatives when they overlook the need for business transformation, Wilson said.
They mistake experimentation for transformation, pushing out multiple pilots or proofs of concept without a clear pathway to scale, he said.
Poor data quality, fragmented ownership, lack of executive sponsorship, unrealistic expectations, and insufficient focus on workforce also can hinder organisations from generating AI returns, he added.
“Organisations that generate the greatest ROI tend to be highly disciplined in prioritising a small number of use cases with clear economic value, establishing measurable success criteria upfront, and scaling proven solutions through governance, change management, and operational integration,” he said.
Governance a necessity, not a cost, to ROI
In fact, governance, monitoring, and security should be seen as necessities, rather than overheads that reduce ROI, as companies bring AI into their environment, Wilson said.
As they deploy increasingly sophisticated AI and agentic solutions, organisations need confidence that outputs remain accurate, reliable, compliant, and aligned with their business objectives, he said.
“Without effective monitoring, organisations can experience model drift, performance degradation, regulatory exposure, or unintended outcomes that quickly erode the value generated,” he explained.
Strong security controls also protect sensitive data, intellectual property, and customer trust, without which organisations cannot scale AI and achieve enterprise-level ROI, he said.
And as companies scale their AI tools and agentic workflows, AI spending will become harder to predict and control.
Gartner expects this to drive a shift towards real-time AI FinOps (financial operations) governance, from retrospective cost reporting.
By 2028, 60% of Global 500 companies will embed AI FinOps control at inference, moving cost governance from reactive reporting to real-time optimisation.
“As AI becomes a larger operational expense, organisations will increasingly focus on measuring cost per task and token efficiency to maintain margins and maximise value,” Gartner said.
It urged organisations to implement runtime cost controls, deploy inference-path telemetry, and make cost governance a core requirement of AI platforms and applications.
Keeping costs under control
Amidst recent headlines on growing token consumption and cost, organisations are realising that AI spending can spiral out of control if they open access indiscriminately to employees.
One CFO in US reportedly chalked up an AI bill of $500 million because the company did not cap its token use, while Uber put a limit on its monthly AI spend per employee after the ride-sharing operator blew out its annual AI budget in just four months.
To have a better hold on their expenditure, Chan recommended companies look at the use cases and outcomes they want to achieve, and prioritise those that generate the most value for their business.
They then can decide the number of tokens needed to support these projects, she said.
Deploying more specialised AI models also will help contain costs and deliver better value for organisations, Dhumatkar said, noting that there will be more of such AI models in future.
Wilson also noted that token cost is just one component of the AI value equation.
More importantly, companies need to determine whether AI is generating measurable business value that justifies the cost of deployment and operations, he said.
“A narrow focus on token costs can lead organisations to optimise for efficiency, while missing larger opportunities for growth, productivity, and innovation,” he noted.
For example, an AI solution that significantly reduces cycle times, improves customer outcomes, enhances decision quality, or enables employees to work at higher-value levels, may generate benefits that outweigh infrastructure costs.
Discussions here should shift from the cost of AI to the value created by AI.
“Organisations should track realised business outcomes, while maintaining discipline around model selection, architecture, and operating costs,” Wilson said.
To ensure their AI efforts lead to tangible ROI, he also urged organisations to establish the operational foundations essential for scale, including data readiness, process redesign, and integration into day-to-day operations.
Responsible AI governance also should be embedded from the start, he said, reiterating the importance of risk management and security frameworks.
He noted that more complexities around these key components, including governance and workforce transformation, will surface in future as AI becomes more pervasive.
A company’s competitive advantage then will depend less on access to AI, and more on how effectively they apply the technology to their business.
And as humans increasingly work alongside AI agents, organisations that deploy without evolving their processes and workforce strategies may struggle to establish sustained returns, Wilson said.
