Companies keep pouring money into AI, but most of that money is not turning into results. A new set of reports from Infosys, Deloitte, MIT and Gartner all point at the same uncomfortable pattern, and together they tell a clearer story than any one of them does alone.
Start with the Infosys numbers. Nearly three out of four senior executives said they have scaled fewer than a quarter of their AI test projects into real, ongoing use. Two out of three said their company cannot even properly measure whether AI is paying off. That is not a small gap. It means most companies are spending money on AI without a reliable way to know if it is working.
A separate study from MIT this year found something even starker: about 95 percent of company AI pilots delivered no measurable financial return, despite an estimated 30 to 40 billion dollars already invested across the industry. Deloitte's research adds another layer, finding that only 1 in 5 senior leaders believe their company is actually ready to redesign how work gets done around AI agents that operate on their own. Companies are buying the technology faster than they are changing how they work, which is exactly why so many pilots stall.
Here is the part that matters most for anyone deciding where to spend an AI budget. Gartner's research found that companies who track the return on every AI project, treat their AI spending like a portfolio of bets, and are willing to shut down the weak ones, saw positive returns on 81 percent of their AI initiatives. That is a massive difference from the industry-wide failure rate, and it shows the problem is not the technology itself. It is discipline.
The choice of project matters just as much as the discipline around it. Gartner found that the most popular AI projects, things like cybersecurity monitoring, threat detection and IT help desk automation, do not deliver the strongest returns. The projects that actually paid off were less exciting: cutting IT costs and managing tech assets more efficiently, generating synthetic data to train other systems, and using AI to write or clean up software code.
There is a pattern here worth sitting with. Companies chase the AI use case that sounds impressive in a board meeting, while the AI use case that actually saves money is often the boring one nobody wants to present. A recent Deloitte survey found close to half of companies are now running more than 30 AI pilots at once, many of them knowing in advance that most will not survive. That is a lot of wasted effort dressed up as ambition.
The takeaway for any business, regardless of industry, is not to slow down on AI. It is to stop treating pilots as free experiments. Pick a narrow, specific problem tied to a real cost, measure it every quarter against a clear number, and be willing to walk away if it is not working after a fair trial. Companies that already do this are seeing returns on the large majority of their AI projects. The ones still chasing headlines are the ones stuck at a quarter of that.