By Adam Mills

AI Editor

0

Share

Newsweek is a Trust Project member

See more of our trusted coverage when you search.Prefer Newsweek on Googleto see more of our trusted coverage when you search.

Register for "How Agentic Operations Turn AI Progress Into Actual Business Value," an upcoming Newsweek webinar, sponsored by Genpact. More details below.

AI agents are increasingly doing tangible work inside large companies, from checking invoices against purchase orders to helping scientists design experiments. But when a task becomes faster or requires less manual intervention, the financial benefit isn’t always obvious, particularly if the broader process and the way employees spend their time remain unchanged.

Read More on AI

A new Genpact study, titled “Autonomy in Practice,” surveyed 2,005 executives across 17 industries, including manufacturing, banking and health care. It found that 79 percent of the executives reported agents in production, yet a weighted average of just 49 percent of AI initiatives funded over the past 24 months had delivered their expected business value. Only 15 percent of respondents said their organizations had fully connected AI investments to specific profit-and-loss outcomes.

Sanjeev Vohra, chief technology and innovation officer at Genpact, believes companies have made substantial progress deploying AI but are struggling to change how work gets done around it.

“Technology is not a problem. In fact, there’s an overdose of technology,” Vohra said in an interview with Newsweek.

He argues that giving employees new tools is relatively straightforward. Restructuring the work those employees perform, particularly when it involves multiple departments or business systems, presents a more complicated challenge.

In its study, Genpact uses the term "autonomy dividend" to describe the value enterprises capture when AI-driven improvements translate into better business performance. The report argues that existing operating structures have become a major obstacle to realizing those benefits.

Elena Christopher, global head of Genpact Intelligence and lead author of the report, said the research was designed to examine whether organizational limitations, rather than shortcomings of the technology alone, were preventing companies from achieving the returns they expected.

One of the clearest disconnects concerns what happens to time saved through AI. While 77 percent of executives surveyed said their organizations had a defined approach to redeploying that capacity, 46 percent cited difficulty putting it to use as a barrier to realizing measurable value, making it the most frequently reported obstacle.

“Freed time is not freed capacity until someone owns where it goes,” Christopher told Newsweek.

The findings reflect executives’ self-reported assessments rather than independently verified financial results.

From Deployment to Transformation

Although 79 percent of executives surveyed reported AI agents in production at their organizations, the extent of deployment varied considerably.

Twenty-seven percent of respondents reported agents running in isolated use cases, while 32 percent described production deployments across some functions. Twenty percent reported broad, consistent use across multiple business functions.

...

Christopher noted that what respondents considered AI agents in production could include enterprise use of chatbots or copilots, rather than systems executing work across departments.

Vohra drew a similar distinction between tools that assist individual employees and AI that changes how entire business processes operate.

Summarizing emails, preparing documents or helping workers complete routine tasks can produce immediate productivity gains. But those improvements may remain confined to the people using the technology.

Christopher described the danger of introducing AI without reconsidering the work itself.

“The cardinal sin of automation and AI is paving the cart path, keeping processes exactly the same and just thinking that you sprinkle technology on there and they will magically change things,” she said.

Vohra illustrated the distinction with the sequence involved in paying a supplier. A purchasing department raises an order, another employee confirms the goods have arrived and a finance team checks the documentation before issuing payment. Automating one step may make one department more efficient, but the payment process still depends on the handoffs between them.

Companies making greater progress, he said, are approaching those workflows from beginning to end, determining where agents can execute work, where humans need to intervene and how information moves between systems.

He described a life sciences company that initially provided AI tools to individual scientists conducting research. After a year, employees were seeing benefits, but the company had not seen the time-to-market improvements it wanted.

“'Let’s look at [the] whole process. Let’s provide an agentic solution to all these scientists,'” Vohra recalled saying when the company reconsidered its approach.

The resulting shared AI agent helped scientists design experimental methods using knowledge from approximately 10,000 existing procedures. Rather than receiving a tool developed separately from their work, the scientists participated in building and training the system.

Across the survey, only 9 percent of executives said their organizations had completed a significant operating model transformation, despite widespread deployment of agents.

Who Is Responsible for Value?

As part of the research, Genpact used an updated version of its AI Maturity Index to assess companies' progress in deploying AI and adapting their operations around it. The index placed 17 percent of respondents in its highest tier, called Leaders, compared with 12 percent in its previous research. The combined share of Leader and Advanced respondents rose from 27 percent in 2025 to 52 percent in 2026.

Among Leaders, 51 percent reported pursuing proactive, enterprise-wide operating model transformation, compared with 17 percent of respondents across all other maturity tiers. Nearly half of the Leaders reported fully connecting AI value to specific P&L lines, compared with just 7 percent of other respondents.

The study also found that 44 percent of Leaders had CEO-led AI transformation mandates, compared with 29 percent in the rest of the sample.

Genpact updated its maturity model for the new study, adding value realization as an assessment criterion, a change that complicates direct comparisons with the previous research.

Christopher put the question of financial accountability more pointedly.

“If you generated value or an outcome from your use of agentic AI but it’s not recorded and doesn’t hit a P&L, is it really value? Can you really recognize it?” she asked.

...

That responsibility can be difficult to assign when the costs and benefits of automation span departments.

Vohra put the relationship between business and technology teams this way: “Tech is a supporting function. Tech is an enabling function. It’s an architecting function,” he said. “But the pull is coming from the business saying, 'I need to go fast.'”

Genpact is applying that principle internally through Client Zero, its program to redesign the company’s own operations around AI agents.

Vohra said Genpact President and CEO Balkrishan “BK” Kalra personally leads the initiative, with executive and functional leaders reviewing progress regularly. Business-unit and functional leaders are expected to set financial and operational targets, while technology teams provide the necessary capabilities.

According to Genpact, Client Zero and related simplification initiatives have generated tens of millions of dollars in operational savings that the company is reinvesting in its business. It now has 180 agents in production across human resources, finance, IT and business operations. Within one of its legal entities, approximately 40 percent of journal entries flow through without manual intervention.

The approach is also shaping Genpact’s client work.

At an electrical supplies distributor processing roughly 3.5 million invoices annually, Genpact introduced agents to handle tasks including extracting invoice data, matching purchase orders with receipts, identifying potential duplicate payments and communicating with suppliers.

After approximately a year, nearly 70 percent of the invoices were being processed by machines with human validation, according to Vohra.

Scaling these workflows also depends on how easily agents can access business systems.

Only 12 percent of respondents reported having full-stack AI architecture capabilities at enterprise scale. Vohra said companies face difficulties involving legacy systems, data access, integration and decisions about whether to build their own technology or purchase existing solutions.

Christopher described the consequences for agents working across these systems.

“They can actually see the problem, identify the problem but they can’t remedy the problem,” she said.

Without a consistent architecture, departments may adopt different platforms and agents that are difficult to connect or govern. The resulting fragmentation can make deployments more expensive while limiting their ability to operate across functions.

The Workforce Question Remains Open

The research also complicates the assumption that deploying AI agents necessarily means reducing headcount.

Among respondents with agents in production, 69 percent reported increasing their agent populations over the previous year. At the same time, 92 percent said their human workforce remained roughly flat or increased.

...

Christopher cautioned against interpreting the companywide figures as proof that individual jobs were unaffected.

At the electrical distributor Vohra described, the invoice-processing operation was using 12 agents and roughly half as many people as a year earlier. Those figures concern staffing within that operation rather than the distributor’s overall workforce.

The study found that 51 percent of Leaders were creating new roles specifically designed for human-agent operations, compared with 22 percent of other enterprises. Such positions can include agent trainers, exception specialists and employees responsible for auditing AI-generated outputs.

“The next question is, 'What will I do?' That’s the biggest question people have,” Vohra said. “You need to talk upfront before you deploy the solution by saying, 'This is what we see the career paths for you in the future.'”

The increasing authority of AI agents raises questions about how companies supervise them. Only 17 percent of executives surveyed said their organizations had governance embedded across the full agent lifecycle.

Across eight decision-making stages, including identifying and framing problems, generating options and escalating exceptions, responses indicating that AI led decisions with human oversight or acted autonomously averaged 40 percent, up from 19 percent in the previous study. Fully autonomous execution averaged 9 percent.

Vohra described the questions he hears from technology leaders responsible for those systems.

“'If agents go wrong, what is to be done? What’s the strategy? How do we execute?'” he said.

With agents communicating among themselves, he raised a related concern. “How do you know which agent has delivered what value and how do you pinpoint any error to rectify something?”

Even among organizations Genpact rates as Leaders, fewer than half have fully connected AI value to specific P&L lines or embedded governance across the full agent lifecycle.

Vohra acknowledged that Genpact is still learning what it takes to make its own transformation deliver the intended results.

“Nobody has a playbook to drive value,” he said.

The challenges raised by Genpact’s research will be the focus of an upcoming Newsweek webinar, titled “How Agentic Operations Turn AI Progress Into Actual Business Value,” and taking place on Thursday, Feb. 4, at 11 a.m. Eastern. The discussion will explore how companies can translate AI investments into measurable results by rethinking how work gets done.

Gabriel Snyder, Newsweek’s executive editor, enterprise, will moderate the conversation with Sanjeev Vohra, Genpact’s chief technology and innovation officer. They’ll examine how businesses redesign workflows, establish financial accountability and prepare employees to work alongside AI agents.

Register to attend live. If you can't make it, register and we'll send you a full video recording of the event afterward.

View 0 Comments

Request Reprint & Licensing

Submit Correction

View Editorial & AI Guidelines