Why R&D Waste Persists Despite Widespread AI Adoption

There's a new whitepaper making the rounds that puts a hard number on something a lot of R&D leaders already suspected but hated saying out loud: over a third of organizations are dumping 25 to 40 percent of their research budgets into projects that never see daylight. Not delayed. Not pivoted. Dead. And the kicker is that this is happening at companies that have already bought into the AI hype, deployed the tools, checked the box.

The waste doesn't announce itself early, either. Nearly half of the teams surveyed say that when a project finally gets killed, it's after burning through more than a million dollars, often during development or late-stage testing, right when the money spent stings the most. That's the expensive way to find out an idea was never going to work. A cheaper way exists, and it's the same one it's always been: figure out the flaws before committing millions, not after.

So why hasn't AI fixed this already? Because most companies pointed their AI budgets at the wrong stage of the pipeline. The tools got handed to people running data analysis and modeling, the execution grunt work, while the actual decision-making, the calls about which projects to greenlight and which to kill, stayed exactly as gut-driven and political as it always was. AI got faster at answering questions nobody had properly framed yet.

The respondents in this report seem to know where the real leverage sits. They point to ideation and early feasibility work as the moment when better intelligence would save the most money, precisely because that's before anyone has spent serious capital. But that's also the messiest, most ambiguous phase of R&D, the one requiring judgment rather than pattern-matching on existing data. It's a much harder problem to hand to a model than