The corporate rush to deploy generative artificial intelligence has encountered a formidable obstacle: the harsh reality of measuring return on investment. Throughout the year, chief financial officers and IT leaders have increasingly hit the pause button on ambitious AI rollouts, shifting the enterprise conversation from speculative hype to rigorous value-derivation. While organizations continue to procure cutting-edge large language models and productivity assistants, determining whether these tools genuinely improve bottom-line business outcomes remains an elusive target for the vast majority of companies.
This metric-driven paralysis stems from a fundamental mismatch between traditional software accounting and the chaotic, probabilistic nature of generative AI. Enterprises are no longer satisfied with vague assertions of efficiency; they demand demonstrable, quantifiable financial benefits. Yet, as organizations grapple with tokenomics, shadow AI deployment, and hidden verification labor, the challenge of proving tangible value has exposed deep structural inefficiencies in how modern corporations evaluate technological investments.
The Evolving Enterprise AI Landscape: From Hype to ROI
To understand the current enterprise reticence, one must examine the rapid evolution of the generative AI boom. Following the widespread public debut of advanced language models, the initial enterprise phase was characterized by a frantic, fear-of-missing-out (FOMO) adoption cycle. Companies rushed to equip developers, marketers, and administrative staff with tools like GitHub Copilot, Claude, and proprietary internal wrappers. Software budgets ballooned as IT departments scrambled to integrate AI into every conceivable workflow.
However, as fiscal year-ends approached, corporate leadership demanded accountability. The CFO’s office, traditionally tasked with scrutinizing enterprise software licenses against productivity gains, found standard metrics ill-equipped for generative AI. Token consumption metrics—tracking the volume of text processed by models—offered no insight into whether the generated output actually drove revenue, reduced operational friction, or saved labor hours.
This friction has created a unique opening for specialized data analytics firms, such as Domo, which are navigating corporate restructuring and acquisition interest from Progress Software while simultaneously rethinking how internal analytics operate in an AI-saturated environment. According to Ben Schein, Chief AI and Analytics Officer at Domo, the current organizational struggle goes far beyond basic cost modeling. It requires an intimate understanding of the hidden operational drag introduced when AI outputs must be verified, corrected, and contextualized by human workers.
Inside the Enterprise: Managing the Multi-Tool Ecosystem
Schein, who has spent over eight years at Domo transitioning from technical evangelism to overseeing product management and internal analytics, has a front-row seat to the daily friction of enterprise AI usage. Across Domo’s own operations, employees deploy a diverse array of AI assistants, including Quad, Domo Copilot, and Cursor-integrated coding models. Reviewing the efficacy of these tools requires continuous cross-departmental auditing to ensure that the cost of deployment aligns with verifiable business outcomes.
One of the most insidious side effects of enterprise AI adoption is the creation of hyper-accelerated silos. Unlike previous technological revolutions where shadow IT developed slowly through unauthorized software purchases, AI supercharges independent experimentation. Teams innovate rapidly, discovering effective prompts and workflow integrations, but these hard-won lessons rarely escape their immediate departments.
Schein highlights this paradox as a central management crisis: how to honor the creative intent of autonomous teams while maintaining centralized visibility and cross-organizational learning. Addressing this requires answering two foundational questions for modern leadership: first, establishing comprehensive visibility into all active AI workflows across the enterprise; and second, rationalizing those disparate tools to ensure consistent access to institutional knowledge. Without this structural cohesion, companies risk funding redundant, uncoordinated experiments that actively compete with one another.
The Hidden Cost of AI: Beyond Tokenomics and Productivity Slop
For months, tech vendors have pitched generative AI primarily as a linear productivity multiplier. However, recent empirical research paints a sobering picture of actual workplace integration. A landmark study conducted by Standard Social Media Lab and BetterUp Labs revealed that roughly 40 percent of generative AI output qualifies as organizational "slop"—low-quality, unverified text or code that fails to meet professional standards.
Crucially, the study calculated that employees spend an average of one hour and fifty-one minutes dealing with just a single instance of this work slop. When translated into labor costs, this cleanup burden averages approximately $186 per employee monthly. When organizations multiply this figure across thousands of knowledge workers, the financial illusion of AI-driven productivity shatters.
Schein points out that traditional metrics fail entirely to capture these downstream expenditures. Measuring only the cost of API calls or token generation ignores the human capital expended by colleagues forced to wade through poorly structured AI outputs. If an AI assistant saves an engineer twenty minutes drafting a preliminary script, but requires a senior architect two hours of meticulous code review and debugging, the net organizational impact is negative. True productivity must account for verification, cross-checking, and error remediation.
Context Over Data: Solving the Clean-In, Garbage-Out Dilemma
The persistent tech industry adage of "garbage in, garbage out" takes on a terrifying new dimension in the era of generative AI. According to Schein, the more insidious threat facing modern enterprises is "clean in, garbage out." Many organizations have spent years and millions of dollars cleansing their data warehouses, assuming that pristine structured data is the sole prerequisite for successful AI deployment.
Yet, because large language models are fundamentally probabilistic, even perfectly curated data inputs can produce erratic, hallucinated, or ungrounded outputs if they lack operational context. The temptation for many executive teams is to endlessly feed more unstructured data into their models, hoping that sheer volume will compensate for a lack of foundational framing.
Domo’s internal analytics strategy takes a different path, prioritizing contextual grounding over raw data accumulation. By auditing legacy dashboards, historical business applications, and institutional metrics that human experts previously deemed vital, organizations can construct custom chat agents tethered to established business logic.
Schein explains that while unstructured data remains critical, structured data—when paired with precise instructions, explicit human guardrails, and historical context—retains immense, untapped value. By bringing historical performance metrics to the forefront through conversational interfaces, companies can anchor AI models in reality before exposing them to the chaotic expanse of uncurated enterprise documents.
The Forever Question: Data Foundations and Multiplied Risk
The struggle to establish reliable data foundations is hardly new. For a quarter-century, enterprise data professionals have battled the persistent disconnect between executive business strategy and the messy realities of database implementation. Schein refers to this as "the forever question"—a perennial cycle of heavy tech vendor spending aimed at fixing data governance issues that inevitably resurface as business models shift.
However, artificial intelligence introduces an unprecedented multiplier effect to this age-old dilemma. In a traditional software environment, a data error or a flawed business logic assumption remains contained until a human operator identifies and corrects it. With autonomous or semi-autonomous AI agents operating at scale, a single flawed assumption or misconfigured workflow can be replicated a thousand times in mere seconds. Because AI operates continuously without human fatigue, the speed and scope of potential corporate damage increase exponentially.
Despite these heightened risks, industry experts emphasize that organizations do not need to wait for a state of utopian data perfection before deploying AI tools. Perfection is an unattainable baseline that paralyzes innovation. Instead, the path forward requires establishing clear deterministic processes, robust verification loops, and transparent business context to safely corral probabilistic models.
Broader Industry Implications and Future Outlook
The current industry-wide pause in enterprise AI rollouts signals a mature phase of technological adoption. The honeymoon period of uncritical enthusiasm has officially concluded, replaced by the demanding scrutiny of the corporate finance department. Vendors who fail to provide transparent pricing models that account for hidden verification labor, downstream cleanup, and context integration will likely find themselves locked out of enterprise budgets.
For organizations navigating this transition, the imperative is clear. Success with generative AI is no longer measured by the number of active licenses distributed or tokens consumed, but by the rigor of internal governance, the clarity of institutional context, and the ability to link AI utilization directly to measurable business outcomes. As companies learn to navigate the fine line between productivity gains and operational slop, the next generation of enterprise AI will belong not to those who deploy the fastest, but to those who measure the smartest.
