Agentic AI is boosting productivity but ROI still lags

AI spending is rising

About one in four dollars spent on AI is wasted, according to a recent Harness report. Yet investment is still moving up. Accenture found that 82% of C-suite leaders plan to increase their AI spending.

The problem is both simply the cost of AI and how companies deploy it. Enterprises can add AI to many teams and processes without changing how work gets done. That may improve individual tasks, but it doesn’t guarantee a measurable business return. More AI usage is not the same as more AI value.

Executives therefore need to manage AI investment against business outcomes. Measures such as the number of pilots, users, models, or AI-enabled applications show activity. They do not, on their own, show economic impact. A stronger case connects AI investment to outcomes such as lower operating costs, shorter processing times, higher revenue, or improved service levels.

This also changes how leaders should think about rising AI budgets. The fact that 82% of C-suite leaders intend to spend more does not prove those investments will pay off. The Harness finding on wasted spending shows the execution risk. Increasing the budget without improving how projects are selected and measured can increase waste along with capability.

The priority is therefore not to slow AI adoption for its own sake. It is to make spending accountable. Each major deployment needs a defined business problem, an expected outcome, and a way to verify the result. That discipline becomes more important as AI moves from relatively contained experiments into core operations.

AI ROI can be difficult to measure when benefits appear across several functions or over a longer period. Early experiments can also have learning value even when they do not produce immediate financial returns. Executives should not classify every unsuccessful pilot as waste. The key distinction is between purposeful experimentation and spending that continues without measurable learning or a credible path to business value.

A few high-impact agentic AI deployments can produce more value than hundreds of shallow pilots

Executives should go deep with AI before going wide. Enterprises should identify a small number of workflows where agentic AI can make a material difference, then develop those systems until their business results can be measured.

Agentic AI goes beyond generating an answer or assisting with a single task. An agentic system can execute multiple steps toward a goal, use tools or data, and make bounded decisions within a workflow. That makes it potentially valuable for processes with repeated decisions and actions. It also raises the need for strong controls because the system can act rather than only advise.

The selection of the workflow matters more than the number of deployments. A company running hundreds of pilots can generate substantial activity while proving little about financial impact. A smaller portfolio gives teams more capacity to integrate AI with existing data, applications, controls, and operating processes. It also makes ownership and measurement clearer.

The business case should start with the workflow. Leaders need to identify where work is expensive, slow, repetitive, or constrained by manual coordination. They can then determine whether an agentic system can remove enough work or delay to justify its implementation and operating costs. The final test is a reportable outcome.

This approach also makes scaling more disciplined. Once a company can show that an agentic workflow works under real operating conditions, it has evidence for deciding where to invest next. Expansion then follows demonstrated value rather than pressure to introduce AI everywhere at once.

Concentration does not mean ignoring experimentation. Companies still need pilots to test technical feasibility, user behavior, security, and economics. The problem is allowing pilots to multiply without clear criteria for stopping, scaling, or measuring them. Agentic systems also require tighter governance than simple AI assistants because autonomous actions can create operational, security, compliance, and financial consequences.

AI is already changing jobs, and companies expect new AI-focused roles to follow

AI-driven workforce change is already underway. Accenture found that 57% of employees say their jobs have changed because of AI. Meanwhile, 78% of leaders expect employee roles to change within the next year. The gap matters. Employees are experiencing the transition now, while many executives still describe it as a near-term development.

For C-suite leaders, the core issue is job design. Adding AI tools without changing responsibilities, processes, or performance measures limits their value. Leaders need to decide which tasks AI should perform, which decisions should remain with people, and where employees need new skills. This requires examining specific workflows rather than treating an entire job as a single unit.

Accenture found that nearly three-quarters of leaders expect their organizations to create new AI-focused entry-level roles over the next few years. This suggests that AI adoption can change the composition of junior work rather than simply reduce it. New employees may enter organizations with responsibility for using, supervising, evaluating, or supporting AI systems.

That creates a practical challenge for talent development. Entry-level work has traditionally helped employees acquire business knowledge through routine tasks. If AI takes over some of those tasks, companies need deliberate ways to develop judgment, domain expertise, and decision-making skills. Hiring for AI familiarity alone will not solve this problem. Employees still need to understand the business processes in which AI operates.

Executives should therefore connect AI strategy with workforce planning. Technology investment, job redesign, recruitment, training, and career development need to move together. The objective is not merely to increase employee AI use. It is to determine how people and AI systems divide work in ways that produce measurable results while preserving appropriate human accountability.

The Accenture findings indicate substantial job change, but they do not establish that AI will produce a net increase or decrease in employment. A changed job is also not the same as an eliminated job. Leaders should distinguish task automation, role redesign, and headcount effects when planning workforce changes. They should also avoid assuming that every function will change at the same rate; the effect depends on the work being performed and the capability of the AI system.

Executives need to use AI themselves if they expect employees to change how they work

Only about one-quarter of top executives use AI every day, according to Ashraf. That creates a clear leadership problem. Companies are increasing AI investment and expecting major workforce change, yet many senior decision-makers do not routinely use the technology they are asking their organizations to adopt.

Executive use matters because AI changes more than individual productivity. Leaders must make decisions about investment, governance, risk, job design, and performance. Regular practical experience can give them a clearer understanding of what current AI systems do well, where they fail, and when human review remains necessary. Without that experience, executives risk setting expectations that do not reflect how the technology behaves in real work.

This does not mean every CEO or CFO needs to become an AI engineer. Senior leaders should instead develop working competence. They should understand how to use relevant AI tools, assess their outputs, protect sensitive information, recognize common failure modes, and know when results require verification. Their own behavior also signals that AI adoption is an operating priority rather than an isolated technology program.

Leadership behavior should be paired with clear rules. Encouraging widespread experimentation without approved tools, data controls, or accountability can create security and compliance risks. Executives need to model responsible use: work within governance policies, protect confidential data, check important outputs, and remain accountable for decisions supported by AI.

Daily executive usage should not become an objective in itself. The real goal is better leadership of an AI-enabled organization. A CEO may have fewer useful daily AI tasks than an executive responsible for software development or customer operations. What matters is that senior leaders understand the technology well enough to make credible decisions about where it belongs and where it does not.

Executive adoption can support cultural change, but it cannot compensate for poor technology, weak data, missing training, or unsuitable workflows. Employees are more likely to use AI when it makes their work meaningfully better and when expectations are clear. Leadership modeling is therefore one part of adoption.

AI creates transformative value only when companies redesign how the business operates

AI that helps employees complete existing tasks faster can improve productivity. But productivity gains alone do not amount to business transformation. Companies need to redesign roles, workflows, data systems, governance, and measurement if they want AI investments to produce broader business value.

The distinction matters for the C-suite. Most early AI adoption has been assistive. Employees use AI to draft content, summarize information, analyze data, write code, or accelerate other existing work. These applications can save time, but the underlying process often remains unchanged. As a result, the company may gain minutes at the task level without significantly improving the cost, speed, quality, or capacity of the complete business process.

Process redesign starts with the desired outcome. Executives should identify where decisions are delayed, where manual handoffs add cost, and where repetitive work can be automated safely. They can then determine which steps should be handled by AI, which require human review, and which can be removed entirely. This shifts the focus from deploying more AI tools to improving measurable operating performance.

Data is a major constraint. AI systems cannot reliably support important workflows when business data is incomplete, inconsistent, inaccessible, or poorly governed. Companies therefore need to improve the data foundation behind high-value AI deployments. This includes deciding which data systems AI can access, maintaining appropriate quality, controlling permissions, and establishing ownership for the information used in automated decisions.

Governance must evolve at the same time. As AI moves from recommending actions to executing parts of workflows, companies need clear accountability for what those systems do. Controls should define permitted actions, escalation rules, human review requirements, security boundaries, and procedures for handling errors. Audit mechanisms should also allow leaders to connect system activity with business outcomes and identify failures.

Measurement completes the operating model. Executives need evidence that AI changes business performance. Appropriate measures depend on the workflow but can include operating cost, cycle time, error rates, revenue, service quality, throughput, or other defined business outcomes. Baselines should be established before major deployment where practical so leaders can distinguish actual improvement from assumed value.

Business redesign does not require changing every process at once. Broad transformation without proven use cases can add cost and operational risk. A stronger approach is to redesign selected workflows where AI has a credible economic case, measure the results, and scale what works. Human oversight should also reflect the consequences of each decision. Low-risk automation can support greater autonomy, while financially material, safety-critical, regulated, or customer-sensitive decisions may require stronger controls.

AI transformation is therefore primarily an operating-model decision. Technology capability matters, but the larger constraint is whether the organization can change processes, responsibilities, data practices, controls, and performance measures around that capability. Companies that leave those elements unchanged are more likely to achieve incremental efficiency than structural improvement.

Main highlights

• Control AI spend through outcomes: About one in four AI dollars is wasted even as 82% of C-suite leaders plan to increase investment. Tie new spending to defined business results rather than adoption or pilot counts.
• Go deep before scaling wide: Prioritize a few agentic AI workflows with clear economic value instead of funding hundreds of shallow pilots. Scale only after deployments produce measurable results under real operating conditions.
• Redesign jobs as AI changes the work: While 78% of leaders expect jobs to change within a year, 57% of employees say that change has already started. Update roles, skills, training, and career paths alongside AI deployment.
• Make AI competence a C-suite requirement: Only about one-quarter of top executives use AI daily, according to Ashraf. Leaders should develop enough practical experience to make informed decisions about AI investment, risk, governance, and adoption.
• Redesign the business: Faster existing work delivers incremental productivity. Rework high-value processes, data foundations, governance, accountability, and performance measures around AI to capture larger gains.