AI ROI is rising but cost savings aren’t driving it

AI is creating more business value

AI now supports 30% of work tasks at the companies surveyed, up from 25% a year ago. Yet the strongest benefits are not lower costs or higher productivity. SAP’s research found that companies are getting more value from AI when employees use it to generate insights, make decisions, and interact with customers.

This changes how executives should evaluate AI. A narrow business case built around labor savings can miss value created elsewhere. Faster analysis can shorten decision cycles. Better access to information can improve the quality of decisions. AI-assisted customer interactions can improve service capacity and responsiveness. These outcomes matter even when they do not immediately reduce operating expenses.

The measurement problem is therefore becoming a management problem. Different parts of an organization can define AI success in different ways. Finance may expect measurable cost reductions. Operations may focus on throughput. Sales and service teams may care about response times, customer outcomes, or revenue. Without an agreed baseline and target, companies can increase AI use while remaining uncertain about its actual return.

Executives should define value at the process level before expanding an AI deployment. The relevant measure could be cost per transaction, decision time, conversion rate, service resolution time, employee output, or another business outcome. AI usage itself is not a return.

SAP’s research found that satisfaction with AI’s value is increasing, while cost efficiency and productivity are not the leading benefits reported by respondents. Kask put the distinction clearly: “It’s not the No. 1 benefit driver of AI, but it’s certainly part of it.”

AI adoption is rising faster than enterprise-wide integration

The next constraint is scale. AI already assists with an average of 30% of tasks among surveyed organizations, compared with 25% last year. Respondents expect that figure to reach 48% within two years. But only 18% of companies report end-to-end, cross-functional AI deployments.

That gap matters. Using AI for individual tasks is much easier than redesigning a complete business process around it. A worker can use AI to summarize a document, draft content, or analyze information without changing the systems around that task. End-to-end deployment is different. It can require AI to work across teams, access multiple data sources, interact with business software, follow permission rules, and produce outputs that other systems or employees can trust.

The figures therefore point to two different forms of adoption. AI use is becoming common at the task level, while deep integration remains uncommon. More employees using AI does not automatically mean the organization has transformed how work gets done.

For executives, this distinction should shape investment decisions. The next stage is not simply to provide more AI tools. Companies need to identify complete processes where AI can produce a measurable business result. They then need the data access, system integration, controls, and operating changes required to deploy it at scale.

This is also why the projected rise from 30% to 48% of AI-assisted tasks should not be treated as proof of higher ROI. Usage measures activity, not economic value. The more important test is whether expanding adoption improves defined business outcomes without creating disproportionate cost, operational risk, or management complexity.

The survey nevertheless shows substantial room for progress. With just 18% of companies reporting end-to-end, cross-functional deployments, most organizations have not yet captured the potential value that could come from redesigning larger workflows rather than optimizing isolated tasks.

AI returns are rising faster than spending, but the economics still need proof

The average U.S. company surveyed spent $37.2 million on AI this year, according to SAP and Oxford Economics. Companies expect that spending to increase by 46% over the next two years. They reported $9.9 million in AI ROI this year and expect that figure to reach $26.5 million over the same period.

The direction is important. Expected ROI is growing much faster than planned spending. Reported ROI would increase by about 168%, from $9.9 million to $26.5 million, while spending would rise 46%. Companies therefore expect existing investments, accumulated experience, and broader deployments to produce more value as AI matures.

This makes measurement discipline critical. AI portfolios should be tied to defined business outcomes and reviewed against a baseline. A company should know what a process costs before AI, what changes after deployment, and which gains can reasonably be attributed to the technology. Revenue effects, cost savings, productivity improvements, implementation expenses, model costs, integration work, and ongoing oversight all affect the economic case.

The spending forecast nevertheless signals strong enterprise commitment. A 46% planned increase implies that companies do not see AI as a short-term experiment. The stronger test comes next: converting that investment into repeatable financial results as deployments move beyond individual tasks.

Data, skills, and governance become harder constraints as AI scales

Scaling AI exposes weaknesses that small deployments can leave hidden. Data, skills, and governance are persistent barriers to ROI. These are especially important as companies adopt AI agents: systems that can perform tasks and take actions for users rather than only generate information.

Kask, whose position and company are not provided in the supplied text, highlighted a specific governance risk. As companies deploy agents, they can discover “shadow agents” or shadow IT agents operating outside normal controls. These systems may access data they should not see or perform actions that cannot be properly audited.

That changes the risk profile. An AI system that generates an inaccurate answer creates one type of problem. An agent with permission to modify records, initiate workflows, or act through enterprise applications can create direct operational and security consequences. Companies therefore need clear identities, access permissions, action limits, audit records, and ownership for deployed agents.

The core constraint is control at scale. More capable AI requires reliable access to company data and systems, but broader access also increases the consequences of weak governance. Executives should treat permissions and auditability as design requirements rather than controls added after deployment. Skills matter for the same reason: employees must understand where AI can act, who is accountable, and when human approval is required.

A KPMG survey found that leaders at more mature companies are shifting their attention toward scaling AI and defining its role in the business. Deployment itself can reveal weak data practices, inadequate controls, and outdated processes. Those findings are useful if management acts on them.

The goal is not to slow adoption. It is to make larger deployments reliable enough to produce sustained returns. Companies that strengthen data management, technical capability, and governance as they scale will be better positioned to move from isolated AI use to controlled, measurable enterprise deployment.

Strong AI returns require process redesign and board-level decisions

AI spending alone will not produce strong returns. The central management task is deciding where AI can change a business process enough to create measurable value. That requires executives to move beyond funding individual tools and focus capital, technical talent, and management attention on a small number of processes with meaningful economic impact.

Board-level AI literacy is an important part of this shift. Boards do not need to understand every technical detail. They do need enough knowledge to challenge investment assumptions, understand material risks, and distinguish broad AI adoption from measurable business performance. This matters because AI competes with other strategic priorities for limited capital and skilled employees.

Process redesign is equally important. Adding AI to an existing workflow may improve individual tasks without addressing the delays, duplicated work, poor data flows, or approval structures that limit overall performance. Companies seeking larger returns need to examine the complete process, decide which steps AI should support or perform, determine where human judgment remains necessary, and update systems and controls accordingly.

Measurement should be defined before deployment. Management needs a baseline for the existing process and a specific outcome for the new one. Depending on the use case, that could include lower operating cost, shorter cycle time, higher revenue, improved customer service, fewer errors, or greater employee capacity. Without this discipline, higher AI usage can be mistaken for higher AI value.

This approach also makes resource allocation more rigorous. Broader findings show that companies expect AI spending to rise sharply, while data, skills, and governance remain constraints. Executives therefore cannot treat every potential AI project as equally valuable. Investment should favor processes where the expected economic benefit justifies the integration work, operating cost, governance requirements, and risk.

Kask, whose position and company are not identified in the supplied text, places this responsibility at the highest level of the organization: “It’s about the allocation of scarce resources, and not just throwing everything at AI, but doing it in a way that maximizes that kind of ROI.” Kask adds: “I think that’s done at board level, frankly, by picking out where you can really change a process in a strategic way, and then measure the outcome.”

The implication for leadership is clear. AI strategy should not be measured by the number of pilots, tools, or employees using the technology. Boards and executive teams should measure whether selected deployments materially improve important business processes. Companies that can make that connection, and demonstrate it with financial and operational data, will have a stronger basis for deciding where AI deserves further investment.

Key highlights

• Measure AI beyond cost savings: AI is creating more value through better insights, decisions, and customer interactions than through direct cost reductions. Leaders should tie each deployment to specific financial or operational outcomes rather than AI usage alone.
• Scale processes: AI now assists 30% of tasks, but only 18% of companies report end-to-end, cross-functional deployments. Focus investment on complete workflows where integration can deliver measurable business impact.
• Demand proof as AI spending grows: U.S. companies surveyed expect AI spending to rise 46% over two years while reported ROI is projected to increase from $9.9 million to $26.5 million. Set clear baselines and attribution methods before committing more capital.
• Build governance into AI deployment: Data access, skills, permissions, and auditability become harder constraints as AI scales, especially with autonomous agents. Establish controls before agents can access sensitive data or take actions in enterprise systems.
• Make AI a strategic resource decision: Strong returns depend on selecting high-value processes and redesigning them around measurable outcomes. Boards should develop enough AI literacy to challenge investments, assess risks, and direct scarce resources toward the strongest opportunities.