Almost three years after OpenAI’s ChatGPT burst into the public consciousness, generative artificial intelligence (GenAI) has become woven into the fabric of modern work.
From executives experimenting with prompts to employees quietly pasting emails into chatbots, AI’s presence feels unavoidable.
But a recent study from MIT’s NANDA Initiative suggests the business case is far shakier than the hype – with a claim that 95 percent of enterprise AI investments have delivered zero measurable return on investment (ROI).
Despite $30–40 billion in enterprise investment into GenAI, this report uncovers a surprising result in that 95 percent of organisations are getting zero return,” the report stated.
“Just five percent of integrated AI pilots are extracting millions in value, while the vast majority remain stuck with no measurable P&L impact.
“This divide does not seem to be driven by model quality or regulation, but seems to be determined by approach.”
The study was based on a “systematic review” of over 300 publicly disclosed AI initiatives, interviews with representatives from 52 organisations, and survey responses from 153 senior leaders collected across four major industry conferences.
Its findings pointed to a familiar problem: too many pilots, too few deployments, and far too little profit-and-loss (P&L) impact.
“Enterprise AI has become the new slideware – boards see innovation via PowerPoint, but P&L tells a different story,” said Tim Banting, Head of Research & Business Intelligence at Techtelligence.
If the vast majority of AI investments yield zero ROI, the uncomfortable truth is that the technology may not be enterprise-ready or the enterprise isn’t ready for the technology.”
When viewed alongside what’s happening inside workplaces, MIT’s study may be telling only half the story.
While companies struggle to push projects past proof-of-concept, workers themselves are adopting AI at scale often without official approval.
This phenomenon, dubbed Shadow AI, could be the real driver of value – one that enterprises have yet to recognise or control.
The Shadow AI Surge
Across industries, employees are already using GenAI in ways that make their jobs easier.
None of these uses appear in official ROI metrics. They are unapproved, untracked, and in some cases discouraged – yet they are ubiquitous.
This bottom-up adoption mirrors the rise of Shadow IT in the 2000s, when employees started using cloud storage and SaaS apps like Dropbox and Slack long before IT departments formally sanctioned them.
Back then, the hidden adoption often revealed unmet needs and eventually forced enterprises to adapt. Shadow AI may be following the same trajectory.
“Shadow AI is exposing where the enterprise stack is broken. Employees aren’t going rogue, they’re patching gaps and solving use-cases that the platform vendors haven’t thought of,” added Banting.
“This is a repeat of Shadow IT, but with higher stakes. CIOs must balance control with enablement, or risk driving AI adoption deeper underground.”
In other words, while enterprises see little ROI from formal projects, workers are already generating productivity gains – they’re just invisible to official accounting.
Opportunity or Risk?
For leaders, Shadow AI represents both a warning sign and a strategic opportunity.
Unapproved AI use raises serious concerns:
Data security: Employees may inadvertently paste sensitive information into public systems, exposing customer or proprietary data.
Compliance: In regulated industries like finance, healthcare, and defense, unsanctioned use could violate data governance laws.
Quality control: Without oversight, employees may rely on inaccurate or hallucinated outputs, introducing risk into decision-making.
These risks echo the early days of cloud adoption, when CIOs feared that employees storing files on Dropbox or Google Drive could create security holes.
At the same time, Shadow AI offers a roadmap for where real value lies. Because employees gravitate toward tools that genuinely help them, Shadow AI reveals the most natural, high-impact use cases. Instead of fighting adoption, leaders can study these behaviours to design sanctioned workflows that balance utility with governance.
“There’s no shortcut to AI ROI. Enterprises need data governance, integration, and real-world feedback from both customers and employees, not just introducing a chatbot on a homepage,” Banting added.
In this sense, Shadow AI may be the real ROI driver of GenAI adoption – but only if enterprises can bring it into the light.
History suggests that enterprises eventually learn to harness hidden adoption.
In the early 2010s, Shadow IT transformed from a threat into a driver of cloud innovation. CIOs who once tried to block services like Dropbox or Slack eventually integrated them, establishing guardrails around security while allowing employees to benefit.
The same arc could play out with AI. Instead of measuring ROI purely from top-down projects, leaders may need to track how employees are already using AI and build sanctioned, secure frameworks around those organic behaviours.
The Bottom Line
The MIT study is right about one thing: the vast majority of enterprise AI investments haven’t delivered measurable ROI. But that doesn’t mean GenAI isn’t delivering value. It simply means the value could be showing up in places companies aren’t measuring.
Shadow AI is already transforming work. Employees are using GenAI tools to accelerate emails, condense information, debug code, and spark creativity – often under the radar of IT and compliance teams. For now, those gains remain invisible in enterprise reporting.
The real challenge for leaders is not whether GenAI can generate ROI, but whether their organizations can catch up with their own employees. The future of enterprise AI may not be built in boardrooms or vendor demos, but at the desks of workers who have already made AI part of their everyday routines.
Until companies learn to recognize and integrate those hidden behaviours, the paradox will remain: billions spent on enterprise AI, with the real transformation happening in the shadows.
