AI Best Used for Early Research and Development
The Pipeline Leaks Before the Product
A study published in March 2026 links three things that rarely share a sentence: human physical well-being, health outcomes, and the state of the planet. Phys.org. The findings are careful, peer-reviewed, and significant. Yet the odds that they become a product are remarkably low. That is not a comment on the science. It is a comment on how research reaches the world.
Something is wrong in the space between discovery and delivery. Laboratories publish. Companies file patents. And then the money runs out, or the priority shifts, or the evidence gets re-examined too late. The 2026 R&D Benchmark Report, presented by IEEE Spectrum and Wiley and based on a survey of more than 200 senior R&D professionals across North America, the UK, and Europe, puts a number on that gap. [1] More than a third of organizations spend 25 to 40 percent of their R&D budget on projects that never reach market. [1] Read that again. Roughly a quarter to two-fifths of those organizations’ R&D budgets, gone — not into a failed product, but into nothing that ever saw a customer.
The problem is not a lack of ideas. It is not even a lack of intelligence. No one in a modern R&D organization complains about a shortage of data. The complaints are about timing: data arrives too late, sits in fragments, and waits in approval chains until the decision becomes irrelevant. The cost of this is precise. Almost half of the respondents estimate losing more than one million dollars every time a late-stage project is canceled. [1] One million dollars, per project, per failure, repeatedly.
For decades, the standard answer has been discipline: stage-gate reviews, portfolio management, formal kill criteria. Those tools helped, but they lean on the same thing that was missing then and is missing now — timely intelligence. A review is only as good as the information it is asked to judge. That is the landscape the conversation about AI should start with. Instead, it usually starts with benchmarks and demos.
The Surprising Pattern in the Data
Here is what the report finds, and it looks like bad news for AI at first. Adoption has surged. CEOs are betting billions on it — the coverage at WebProNews describes leaders who treat AI as existential for survival, despite high costs and implementation hurdles. WebProNews. Yet the waste persists. The share of budget lost to projects that never reach market has not collapsed. The late-stage cancellations continue. The obvious conclusion would be that AI is another expensive promise that failed to deliver.
That conclusion is too fast. The data does not indict AI; it indicates where AI is being pointed. Most organizations use the technology primarily for execution. They use it to draft, to code, to summarize, to accelerate the tasks that happen after a project has already been approved and funded. That is useful, but it is also late. By the time execution tools are running, the expensive decisions have already been made — the decisions about which projects deserve investment, which assumptions deserve trust, and which ideas should be killed while they are still cheap to kill.
The counter-perspective that completes the picture: the intelligence for consequential decisions is precisely what many organizations still lack. Fragmented data. Lengthy approval processes. Critical insights that arrive too late to influence the investment choice. AI is being used as a better hammer for a process that needed a different frame. The tool is not overrated. It is misplaced.
That is the concrete finding that should change how leaders think. The gain is not in making researchers faster. The gain is in making investment decisions smarter, before the money is committed.
The Gain, Shown Without Hype
Where AI genuinely lifts the work is at the front end, during ideation and feasibility. This is the phase before significant resources are committed, and it is exactly where better intelligence produces the largest return. Competitive intelligence is available earlier. Market signals are woven into the assessment. Patent landscapes — the legal and technical terrain a new product must survive — can be mapped before a single pilot is funded. The report says it plainly: earlier access to this intelligence delivers the greatest impact at ideation and feasibility, before the costly commitments begin.
Consider what that means in practice. A team at the feasibility stage evaluates a new material, a new molecule, a new sensor. The decision to proceed commits the organization to years of development and the millions that go with it. If the competitive intelligence shows that three rivals already hold the key patents, or that the market has shifted, that knowledge is worth every dollar it cost to obtain. The project is killed early, at perhaps ten thousand dollars, not late, at a million. The budget that was saved is real, and it is available for the next idea. That is the mechanism by which R&D waste becomes R&D capacity. The one-million-dollar loss becomes a one-million-dollar investment elsewhere.
This is not a promise. It is a shift of the decision point. The same AI models can be more useful when asked a narrower question: what information would change this decision, and do we have it now? That is a question a machine can answer with evidence, and it is a question that pays immediately. The gain is concrete because the failure it prevents is concrete.
The report’s sponsor, Patsnap, built its business on this exact premise — AI agents and analytics for IP and R&D teams, trusted by more than 15,000 enterprises, law firms, and research institutions. The premise is not that AI dreams up breakthroughs. It is that the landscape around a decision can be mapped completely and in time, and that map saves money and accelerates time to market. The numbers in the report — the 40 percent waste ceiling, the million-dollar late failures — are the negative space that defines the positive gain.
The Consequence Nobody Wants to Draw
Follow the logic one step further, and the picture changes entirely. The organization that spends 25 to 40 percent of its R&D budget on projects that never reach market is not only losing money. It is silently defunding the alternatives. Every late-stage cancellation is not just a loss; it is a forgone breakthrough, a project that was never attempted because the money and the attention were tied up in a failing one. The waste is not an inefficiency. It is a tax on the actual innovation that an organization could have produced.
That is the final consequence no one draws, because it leads somewhere uncomfortable. If the innovation pipeline is leaking billions, then the scarcity that executives constantly cite — too few good ideas, too little budget, too much risk — is partly self-inflicted. The ideas are there. The budget is there. It is simply consumed by decisions that were made without the right intelligence at the right time. And that is fixable with tools that already exist.
The health-planet research from March 2026 is a useful lens. The scientists did their job. They established a link that matters. Whether it becomes a diagnostic, a therapy, or a policy instrument depends on someone making a well-timed bet — and that bet depends on intelligence about competitors, markets, and patents that is neither mysterious nor expensive. The gap between publication and product is not a gap in knowledge. It is a gap in decision-making.
This is where AI actually lifts, and the lifting is measurable. Not in demos. Not in benchmarks. In the quiet act of killing a bad project early, in the millions that stay in the budget, in the alternative product that gets funded because a dollar was not wasted. The most uncomfortable part is that many organizations will not recognize themselves in this description, because their metrics — projects completed, papers published, features shipped — never record the projects that should have been started instead. The leaders who point AI at the decision, instead of at the typing, will not make headlines. They will make the products that other companies cannot afford to attempt. The rest will keep publishing, keep paying, and keep wondering where the innovation went.
Sources
2. Wiley
3. Patsnap