A widely cited MIT statistic found that 95% of AI pilots fail to generate a positive return on investment despite significant financial commitments. The figure has been disputed in some quarters, but the pattern it describes has not. Companies are spending more on artificial intelligence than ever before, and enterprise-level results remain uneven.
The instinct when a technology investment underperforms is to blame the technology. In most failed AI implementations, that instinct leads organizations in the wrong direction. The gap between what AI projects promise and what they deliver is less often a product limitation and more often a failure of preparation: a failure to examine what would actually have to change for the investment to work.
Also read: Goldman Sachs sees significant shift in AI labour market
Gopal Subramaniam, Senior Vice President at Dell Technologies (DELL), has spent years leading large-scale go-to-market and revenue operations teams across a large global organization. He has watched companies pour resources into AI tools only to see adoption stall, workflows stay unchanged and results fail to materialize. In his view, the technology is rarely the problem. The foundational work that should precede it almost always is.
“AI doesn’t pay off when you insert it into today’s workflow,” Subramaniam told TheStreet. “It pays off when you redesign the workflow around what AI can do.”
The real challenges behind tech transformations
Most failed AI deployments trace back to the same decision made too early: choosing a tool before defining the problem it is meant to solve. An organization identifies a capability, builds an implementation plan around it and begins rollout before asking the harder question: what would actually have to change across the business for this to deliver?
Research from RAND on AI project failures points to models that do not fit actual business workflows and tools deployed without solving meaningful problems for the people who use them, alongside weak data infrastructure, as the most common causes. These are not technology failures. They are planning failures that form before a single prompt is run.
A tool layered onto an existing process preserves most of the friction in that process while adding a new layer of adoption overhead. Redesigning the workflow first means the tool has something to deliver into. Without that work, the technology is solving for a version of the job that is already outdated.
“If the goal is to change how work gets done, roles and skills have to change with it,” Subramaniam said. “That includes leadership. Leaders need to be more collaborative, but they still own the decision, even when AI informs it. Too many companies bolt AI onto the existing workflow and change nothing else. An employee might save a few minutes on one task, but if the workflow around it is unchanged, the organization’s output doesn’t move.”
What happens when these areas are ignored?
When organizations treat AI as something to add rather than something to reorganize around, the results are predictable. Productivity gains stay local. The business spends money, employees use a new interface and nothing of consequence changes at the scale that justifies the investment.
That frustration has a second-order effect that is now well documented. A Deloitte survey of workers in the U.K. found that roughly one-third of employees use generative AI without their employer’s approval or knowledge, and half of AI users had received no training from their employer at all. When official tools do not fit actual workflows, employees find their own. The result is an ungoverned spread of AI use that is nearly impossible to audit.
That spread carries real risk. Confidential information entered into an unsanctioned tool may be retained, used in model training or exposed in ways that fall outside a company’s compliance controls. In industries with strict data-handling requirements, a single incident can carry regulatory consequences that far outweigh the cost of the AI project itself.
A smarter business transformation approach
Organizations that extract consistent value from AI investments tend to start from a different place. Instead of selecting a tool and building a case for it, they identify a specific, high-value business problem and work backward to the technology and workflow changes that would solve it. That sequence changes who gets involved, what gets measured and how quickly adoption takes hold.
“Start with a defined business problem, not a tool,” Subramaniam added. “The problem narrows which AI you need and which workflows have to be redesigned. Then bring the people who do the work into that process early. They’ll see how their roles change, and they’ll tell you what it takes for the AI to deliver.”
More Artificial Intelligence:
- AI’s biggest business opportunity may not look like AI
- AI racks are outgrowing old power: 2 chipmakers have a strategy
- BofA splits software into winners and losers as AI agents arrive
That approach also yields more accurate measurements. When the goal is a business outcome rather than a deployment milestone, organizations can track metrics that actually indicate progress. Tool adoption rates measure a behavior. The right business problem frames metrics that measure a result: throughput, error rate, decision speed, cost per unit. That clarity makes it easier to spot where an implementation is falling short and to adjust before the window for course correction closes.
Clear goals also shift how teams respond when results lag. A deployment milestone is binary: it either shipped or it did not. A business outcome has degrees, which means feedback from real workflows can inform adjustments in tools, roles and processes rather than triggering a full reset.
Setting investments up for success
The companies making measurable progress on AI are not uniformly the ones with the largest budgets or the most sophisticated infrastructure. Many are doing less glamorous work that comes earlier: defining the problem precisely, mapping what the process change would look like and treating the people who execute the work as essential to the design process rather than an afterthought to it.
That discipline is harder to put in a press release than a model upgrade or a deployment number. It does not photograph well at a product launch. But it is what separates an AI investment that scales across an organization from one that generates a case study on what to avoid next time.
As AI capability continues to advance, the constraint on enterprise returns is shifting further from the technology and toward the organization around it. The companies that close that gap first will not be the ones that founded better technology. They will be the ones that asked harder questions before they deployed it.
Related: Wall Street just got a candid look at who is winning the AI race
