Why Most Companies Still Get No Return From AI

I keep thinking about what happened right after Watson won Jeopardy.

Everyone remembers the win, but almost nobody remembers what came right after it. Every executive I talked to suddenly wanted a Watson of their own.

They didn't actually want intelligence. They wanted magic.

Building real value with Watson meant redesigning processes, cleaning up years of messy data, redefining who made which decisions, and changing how people worked day to day. That part was hard, so most companies skipped it. When the results didn't show up, they blamed the AI.

I'm watching the same pattern play out right now with generative AI.

Executives tell me they deployed AI. What most of them actually did was drop a chatbot on top of a broken process and expect the process to fix itself on its own. The technology didn't fail here. Leadership skipped the transformation work that makes the technology worth anything.

AI ROI Is Failing Because Companies Start in the Wrong Place

A study from MIT's Project NANDA found that 95% of organizations investing in generative AI report no measurable return. That's against $30 to $40 billion in enterprise spending. Researchers reviewed 300 public AI deployments, interviewed 150 executives, and surveyed 350 employees to reach that number.

I wasn't surprised by it. I watched an earlier version of this exact story unfold at IBM.

The mistake being repeated right now is a sequencing mistake. People still believe AI creates value on its own. It doesn't. AI reveals where value already exists in a business, waiting to be unlocked.

Most organizations buy the model first and ask where the value is later. That order is backwards. What actually works is business value first, then decision readiness, then workflow redesign, and only then AI.

Companies Are Spending AI Budgets on the Wrong Problems

The MIT researchers found something specific worth sitting with. More than half of generative AI budgets went to sales and marketing, the two areas with the most visibility and the least reliable payback. Back office work, the unglamorous stuff like claims processing or contract review, delivered faster and larger returns, and got a fraction of the investment.

I saw the same instinct at IBM. Executives wanted Watson working on the flashy problem, the one their board would notice in a meeting. They rarely wanted it on the boring problem, the one that would have actually paid for itself within a year.

The build vs buy pattern matters too. External partnerships in the MIT sample succeeded roughly twice as often as internal builds. Companies kept trying to construct their own version of something a partner already had running. That instinct comes from wanting control. It usually costs more than it saves.

I ask leaders one question before I look at their AI roadmap. If we removed the AI tool completely, would the underlying decision still be broken? Most of the time, the honest answer is yes. That tells me the AI was never the fix. It was a distraction from the fix.

AI Will Create Value When It Becomes Invisible Infrastructure

Watson taught me something most people have forgotten. Breakthrough technology doesn't disappear once the hype fades; it becomes infrastructure.

Nobody talks about cloud computing anymore, not the way they did back in 2010. Nobody talks about databases or APIs either. Those technologies won so completely that they became invisible, woven into how every company runs.

Generative AI is on the same path. The companies that come out ahead in five years won't be the ones with the flashiest chatbot today. They'll be the ones that rebuilt how work actually happens around what AI makes possible.

AI isn't the product, but the operating system running underneath every product a company builds from here forward.

What Leaders Must Fix Before Investing More in AI

If you're the one signing off on AI spend this year, start with a different question. Not what can AI do for us, but where is value trapped in our business right now, and what's actually stopping us from reaching it.

Answer that honestly, and the workflow redesign becomes obvious. Skip it, and you'll be back here next year, explaining the same zero return to the same board, wondering why the technology didn't work.

Watson taught me this lesson the hard way. I'd rather other leaders learn it from a report than from their own bill.

I'd also tell that leader to be honest about who owns decision readiness inside their company. It's rarely the CIO alone, and it's almost never a vendor. It sits with whoever owns the process that AI is supposed to improve. If nobody can name that person, that's the first thing to fix, well before the next model gets evaluated.

None of this means the technology is overhyped. Generative AI is genuinely capable, more capable than Watson ever was in its Jeopardy days. The gap is in how few organizations have done the unglamorous work required to let it do the work.

AI reveals which organizations mistake technology deployment for business transformation. That's been the whole story since Jeopardy, and it's the whole story now.

← Back to all articles