Wharton Survey of 1,400 Employees: Corporate AI Investment Surges, But Only 27% Track How Much Money It Generates

TechFlow Summary:Whether AI capital expenditures can sustain themselves ultimately hinges on whether downstream companies can turn AI into revenue. This Wharton survey reveals that most firms are still measuring returns by "hours saved," with the share expecting a "significant increase" in next year's budgets dropping from 19% to 10%. The latter section, which covers Workday following Salesforce's lead in offering a "headless" mode, serves as another recent example of fears that SaaS will be reduced to a mere pipeline by AI agents.

The trillion-dollar question amid the AI boom is whether corporate investments in this technology actually yield a return. Yet measuring such returns is often uncertain, confusing, or incomplete.

This is because, as we have written before, AI requires time to generate investment returns that can materialize on a company's income statement.

A new study from the University of Pennsylvania's Wharton School and consulting firm GBK Collective echoes this view: when tracking AI returns on investment, most companies are still measuring whether it makes employees more efficient, rather than whether it generates more revenue. Based on an online survey of 1,400 U.S. employees at companies with at least 1,000 staff and annual revenues exceeding $50 million, the results show that 40% of managers measure AI's business impact by "hours saved per employee," while 37% track how many employee errors AI has reduced.

Only about 27% of managers said they track revenue generated by AI features, apps, and tools offered to customers.

Perhaps even more concerning is that 44% of managers stated that the primary role of AI over the next one to two years will be to boost productivity, while only 8% believe its main purpose is to drive new revenue or create new products.

Of course, boosting productivity is no small feat, especially when it implies companies are using AI to replace labor in roles such as external contractors, customer service, or IT support ticket handling.

However, Jeremy Korst, one of the authors of the Wharton study, noted that it is "very difficult" to find "real, materially quantifiable financial impacts" from AI spending.

He said that measuring returns nowadays relies more on intuition and managers' "perception" of returns rather than hard data. He added that since AI is still in its early adoption phase, this level of return is "normal" and aligns with historical technology adoption cycles.

About 80% of managers in the study reported achieving positive returns from using AI, many of whom work in IT, finance, or human resources departments.

But Korst noted that the 28% of managers who went further to say their organizations achieved a "strongly positive" return were also more likely to be in companies that measure AI returns based on "growth outcomes" (such as newly created revenue) rather than just cost savings from improved employee productivity. (In contrast, a recent survey found that 35% of The Information subscribers—who often come from tech companies and investment firms—reported getting "multiple times" their return from AI spending, though overall only 52% of respondents said they saw a positive return.)

In our own reporting, examples of non-tech companies seeing revenue growth from AI seem few and far between. Novo Nordisk previously told us it built AI agents for auditing and coordinating clinical trials, which it expects to generate tens to hundreds of millions in additional annual revenue. Walmart CEO David Guggina stated at Goldman Sachs' Technology Conference in September that customers using its AI assistant, Sparky, have average order values 40% higher. Most other examples cited in this column over the years have demonstrated savings in manpower and time.

The Wharton research suggests that corporate spending on AI will only continue to grow rapidly if companies prove that these expenditures yield strong returns and tie them to growth outcomes—such as new revenue from AI and higher output (e.g., how much equipment manufacturers produce)—rather than linking them to cost reduction.

Nevertheless, for now, spending continues to grow at a fairly rapid pace. About 45% of managers stated their companies will invest $10 million or more in AI this year, up from 39% in 2025.

However, the research indicates that significant increases in AI budgets by 2027 may not be as widespread. Only 10% of respondents in the survey expect their companies' AI spending to see a "substantial" increase over the next 12 months, down from 19% in 2025.

One bright spot is that companies may be getting better at tracking AI expenditures. About 40% of managers in the survey expressed they were "very" confident in their companies' ability to track such spending, while another 39% said they were "fairly confident." Only 17% said they had "some confidence."

Workday Follows Salesforce, Offering "Headless" Option

On Tuesday, Workday released new tools allowing customers to more effectively use AI agents developed by companies like Anthropic, Microsoft, and OpenAI to automate tasks within their Workday accounts.

These tools are similar to Salesforce's SIforce, which was called AIforce last month and earlier referred to as Headless 360.

As I reported in detail last week, some employees and investors fear that the new tools from Workday and Salesforce actually amount to capitulation to a new wave of AI providers, potentially reducing these software companies to "dumb" infrastructure or merely beautified databases. However, leadership at both Workday and Salesforce has stated they expect to make significant money through this approach.

"I would basically say we're giving customers a choice," CEO Aneel Bhusri said during a media briefing ahead of the company's Las Vegas customer conference. "We offer a world-class user experience directly through Workday, via Sana (Workday's AI platform). But if customers want to choose the headless mode"—meaning using another company's AI to access Workday—"that's fine too," he said.

"I always prefer they use our interface, but the headless mode works just as well for Workday."

Starting in February 2027, Workday will begin charging customers based on the frequency with which AI agents invoke connectors to perform actions within Workday, marking the end of the "grace period" during which the company currently offers such tools for free.

Join the official Coincamps community:

X: https://x.com/coincamps

Telegram: https://t.me/coin_camps