Why AI productivity gains do not always deliver enterprise ROI

AI can make an employee two hours faster and still produce little financial return for the company. For CIOs and CTOs, the decisive question is what happens to those two hours afterward: whether they create capacity, improve service, move people toward more valuable work, reduce costs or contribute to growth. TechTarget asked eight CIOs, technology leaders and AI experts whether AI is paying off, and their answers point to a distinction that matters when companies decide what to scale. Employee productivity shows an effect; enterprise ROI depends on capturing it.

AI can save time without creating enterprise ROI

That distinction changes the question technology leaders need to ask about AI. Software development and sales automation already produce targeted returns, but a successful demonstration can still be a poor candidate for broader investment when its value cannot support its costs. Companies evaluating the next stage of AI spending need to establish how a working application produces an organizational outcome. The debate has shifted from technical effectiveness to the value enterprises can capture from it.

The gap remains visible even when operational results look good. Manish Jain, founder and CEO of Strategic Horizon Research, says AI is improving productivity, customer experience and business-process efficiency, while enterprise-wide financial returns remain much harder to realize and demonstrate. He calls the return concept “ROAI” and says the payoff exists for a relatively small group of companies; for most, it is “not at scale.” Because Strategic Horizon Research operates in the research market around these technology decisions, its business benefits from enterprise demand for such analysis, giving Jain a commercial stake in the discussion.

Brian Jackson, principal research director at Info-Tech Research Group, identifies a related measurement problem. In an Info-Tech survey, 42% of organizations reported department-wide AI adoption with measurable impact, while another 28% had reached department-wide adoption without clear impact. Info-Tech also has a commercial interest in organizations seeking technology research and guidance, so its findings reflect the perspective of a firm serving that market. The 28% result makes the problem concrete because department-wide deployment can exist before an organization establishes measurable business value.

For leaders making scaling decisions, productivity is the beginning of the ROI inquiry. A company may need to redesign jobs, move employees toward higher-value activities or use newly available capacity to serve more customers. Other applications can connect more directly to a valuable outcome and require less change to existing roles. In either case, the organization has to follow the effect beyond faster work and measure what happens to the capacity AI creates.

The missing step is converting efficiency into an outcome

Following that capacity leads directly into existing workflows. Matt Watkins, CIO of IMA Financial Group, sees AI automating tasks, improving work products and preparing associates more effectively for client interactions, but he says the harder problem is translating those gains into meaningful savings and growth. Doing so requires understanding how work happens now and redesigning it around the capabilities AI introduces. A local improvement creates enterprise value when it changes how organizational resources are ultimately used.

IMA’s policy-checking and quote-comparison work shows why that conversion can be difficult. The company has automated much of those processes, yet the automated activities are fragments spread across many jobs rather than complete jobs that disappear after automation. Savings consequently appear as small amounts of time distributed across many people. IMA has to quantify those increments before it can determine whether the newly available capacity is substantial enough to use elsewhere.

Once IMA identifies that capacity, the next decision is where employees should apply it. Watkins asks whether people can perform work that was previously neglected, whether roles should shift toward higher-value and client-facing activities, or whether resources should move to another process. Each option connects saved time to an organizational consequence rather than assuming efficiency automatically creates one. As Watkins puts it: “Without that work redesign, everyone just becomes more efficient in their current work, but we don’t repurpose those hours to something that drives more value for the company and our clients.”

The scattered nature of those savings also explains why individual productivity metrics can conceal the enterprise result. If ten employees each complete parts of their jobs more quickly but continue performing the same set of activities, task efficiency improves while the organization may gain little usable capacity. Watkins’s approach follows the saved time into the subsequent allocation of work. Work redesign matters especially when automation removes pieces of roles while leaving the rest of each role intact.

That workflow test becomes stricter once costs enter the calculation. Chris Campbell, CIO of DeVry University, distinguishes faster task completion from better operational economics because saved time has to be valued alongside what happens afterward. Human effort spent checking AI output belongs in the calculation, as does the cost of operating the technology. “A faster task by itself does not tell me that,” Campbell says of whether AI has delivered the return he wants to measure.

DeVry’s student-advising work gives Campbell’s test a concrete sequence. If AI reduces routine advising work, an advisor can spend more time with a student who needs help; DeVry can first ask whether AI genuinely created that capacity and then whether the additional attention improved service. The second measurement connects the efficiency gain to a service outcome. Campbell is comfortable pursuing such practical gains while requiring better resulting work and economics that still make sense after full costs are included.

Those workflow tests also explain Jain’s broader “ROAI” argument. Companies can count usage, experiments and adoption, but those measures establish financial value only when their effects continue into added capacity, better service, higher-value work, savings or growth. The resulting benefit must also support the costs of checking AI’s work and running the underlying technology. Value realization, in Jain’s terms, means carrying an observed AI effect through to that organizational result.

The workflow examples also set a boundary around work redesign as a mechanism. Redesign is especially useful when savings are dispersed through existing roles, as they are in IMA’s policy and quote workflows. Applications tied more tightly to revenue, customer experience or a bounded operational process can have a shorter path from AI use to measurable value. Companies can choose the mechanism that fits the workflow while still requiring a credible connection between efficiency and outcome.

Some use cases shorten the conversion path

That shorter path shapes how Tony Garcia, chief information and security officer at Infineo, controls AI consumption. Everyone in his environment receives an AI usage limit, and users are expected to report how much time AI saved and identify the associated business value or benefit. Infineo declines to pursue uses when users cannot demonstrate that benefit. Garcia contrasts that discipline with indiscriminate AI consumption, which he calls “tokenmaxxing nonsense.”

Sales is one of Garcia’s examples because the outcome sits close to the AI-assisted work. A branding skill can allow a sales team sitting down with a customer to produce a fully branded sales deck by the following day; Garcia notes that Anthropic also has such a branding skill. That faster turnaround connects directly to the sales cycle and customer experience. The organization can consequently assess the value inside a specific commercial workflow instead of depending on a broad reallocation of employee time.

Infineo’s “AI code factory” provides a second bounded workflow. Garcia says many other organizations are building code factories as well, and Infineo uses agents that write code, audit it against company standards, conduct the security review and prepare that review for people to inspect and approve. Human approval remains part of the process, while AI changes several linked stages leading to that approval. Development and security value can then be evaluated against the work performed across the sequence.

Garcia’s vending-machine example clarifies the selection rule behind those investments. A consultant could use AI to run a vending machine, but adding AI to a process does not by itself make the process worth funding. In Garcia’s view, the sales and code-factory applications have demonstrable business benefits and “meaningful impact,” while there are also “nonsense use cases.” Infineo’s operational rule is to demand evidence of benefit before increasing AI consumption.

A similarly selective approach appears at ARG. Jim Begley, CTO of ARG, says the company describes itself as a “fast follower”: it watches what is happening in the market and looks for point-based requirements where AI can generate a net-positive impact. ARG assesses the intended result strategically, then follows that assessment with strategic implementation. Begley says the company has observed positive ROI from that approach, while ARG’s commercial position in the technology market gives it a stake in how enterprises evaluate and implement technology investments.

Begley also limits the scope of that ROI claim. “That’s not a global answer,” he says, adding that some investments have been made without appropriate ROI evaluation and that some things have no ROI. His narrower claim is that careful assessment of the intended outcome and equally careful implementation can produce positive ROI. That distinction adds a second route beside broad work redesign: bounded applications can connect operational or commercial work to measurable results over a shorter chain.

Strategy and operational discipline raise the odds of measurable impact

Use-case selection still operates within a wider enterprise environment, and Info-Tech’s results show a strong association between strategy maturity and measurable impact. Organizations with a dedicated AI strategy approved by their boards reported substantially higher levels of department-wide adoption with measurable impact than organizations still drafting a strategy. Organizations that partially incorporated AI goals into wider IT or digital strategy fell between those groups. The progression appears across all three strategy states.

Those Info-Tech figures show an association, so they do not establish strategy as the cause of the difference. Jackson nevertheless recommends formalizing a dedicated AI strategy where possible and incorporating AI into a broader IT or digital strategy where a standalone strategy is impractical. A defined strategy gives the organization a place to specify desired outcomes and coordinate implementation around them. That coordination matters because adoption can otherwise expand faster than the organization’s ability to account for its impact.

Strategy then has to become an operating practice. Eric Helmer, executive vice president and CTO of Rimini Street, says AI is paying off at a smaller scale than many expected because organizations face difficulties “operationalizing AI responsibly.” Governance frameworks, security controls, data readiness, regulatory compliance and clear accountability are often developing more slowly than the technology itself. Rimini Street has a commercial stake in enterprise technology services, so Helmer’s assessment also reflects the perspective of a company that can benefit from demand for the operational work he describes.

Those operating conditions include people as well as technical controls. Helmer says employees can remain uncertain about how AI will affect their roles, which can slow adoption, while stronger success stories come from companies that address trust, transparency and organizational change alongside the technology. Those concerns affect whether AI becomes part of routine work and whether organizations can rely on the resulting processes. Governance and change management consequently affect value realization because sustainable deployment depends on both.

For CIOs and CISOs, those controls become part of the economics of scaling. Security, data readiness, compliance and accountability consume effort, while weak foundations can keep a working AI application from becoming a dependable enterprise process. Responsible operationalization means establishing enough control to sustain and measure the intended outcome. Scaling a model or tool before those conditions are ready can increase deployment faster than realized value.

Low-level usefulness may still fail the enterprise economics test

The economics become more demanding as AI spending rises because genuine usefulness can still produce too little value to support the services providing it. Futurist and keynote speaker Nikolas Badminton separates the value people receive from large language models, or LLMs, from the return sought by big technology companies and frontier labs, the companies developing the most advanced general-purpose AI models. His concern is whether commonplace improvements to everyday work can bear much higher costs. That question shifts the ROI test from usefulness alone to usefulness at a given price.

Badminton bases that view partly on a council of 50 people for a software company where participants discussed their AI use. Frequently mentioned applications involved systems learning how a person works and acting as a personal assistant, along with helping to write emails and documents. He expects this kind of low-level work to become standard in “the next year or two.” As those uses become routine, their presence alone gives an enterprise little basis for accepting substantially higher AI spending.

That cost question becomes sharper in Badminton’s prospective pricing scenario. He points to companies that have had “trillions of dollars” invested in them and contrasts a “$100 a month” subscription with a possible future request for “$1,000 or $2,000 per seat” as providers seek investor returns. Those higher figures describe a prospective scenario rather than an observed market-wide pricing change. Badminton’s argument is that modest improvements to low-level work would struggle to justify costs at that level.

The pricing scenario brings Garcia’s consumption discipline back into the economic calculation. A use case may save time and employees may genuinely prefer working with it, while the resulting benefit can still be too small to support the technology bill at enterprise scale. Changes in cost can also move the same use case from a plausible investment to a weak one. ROI evaluation therefore has to keep operating cost attached to the outcome the organization is buying.

Key takeaways for leaders

• Connect productivity to enterprise ROI: CIOs need to trace AI time savings into added capacity, better service, higher-value work, lower costs or growth. Productivity metrics alone do not establish a financial return.
• Redesign work around AI capacity: Organizations with AI savings spread across many roles need to quantify that capacity and decide where to redeploy it. ROI calculations should also include human review and technology operating costs.
• Prioritize bounded use cases: AI applications in areas such as sales and software development can offer a shorter path to measurable value. Technology teams can require a defined business benefit before increasing AI consumption.
• Build strategy and operating discipline: Organizations with more mature AI strategies report higher measurable impact, although the data shows association rather than causation. CIOs and CISOs need governance, security, data readiness, compliance and accountability that support sustainable deployment.
• Test usefulness against full economics: Everyday AI assistance may save employees time while producing too little value to support rising technology costs. Buyers need to reassess use cases as pricing changes and compare total operating costs with measurable business outcomes.