AI won't have your next big idea. Here's the part of innovation it actually transforms.

You put AI into your innovation process expecting a fountain of new ideas, and what came back was more mixed than that. In some places it was transformative, compressing weeks of prototyping and testing into days. In others it fell flat, offering neither the breakthrough idea you hoped for nor much help with the nervy judgement of when to launch. It is easy to read that as doing it wrong. You were not. You were simply pointing AI everywhere at once, when the evidence says it earns its keep somewhere specific.

Here is the map. A 2025 systematic review in the journal Systems, synthesising 103 high-quality studies, measured where AI actually lands in innovation, and the pattern is precise. AI is used most heavily, and most successfully, in the development of new products, scoring 4.48 out of 5, noticeably higher than in idea generation (4.28) or commercialisation (4.34). And the trend is steep: AI's use in new product development rose from 13% to 24% in a single year. Read together, that says something you can act on. AI is remarkable at developing, testing and refining ideas, and weaker at having them or knowing when to release them. So stop spreading it evenly, and put it where it compounds.

Where in innovation does AI actually earn its keep?

In the messy middle: development, prototyping, testing, refinement. The review's number is unusually specific, with AI use peaking at the development stage, and the real-world cases make it concrete. General Motors uses AI to generate conceptual vehicle designs. Unilever uses it to develop novel cleaning-product ingredients. Nike built a virtual platform, Nikeland, to test product concepts and gather consumer data in real time. DeepMind and the UK's National Health Service built a system that detects acute kidney injury up to 48 hours earlier than before. In every case AI is accelerating the build-and-refine loop, not supplying the original spark.

This is the same shape as designing the loop rather than running each step: AI multiplies the iterative, effortful middle of a process, where the work is largely about trying, measuring and improving. Point it there and a small team can explore far more possibilities than it ever could by hand.

So what stays human, the idea and the launch?

Largely, yes, and it is worth being deliberate about it. AI scores lowest in innovation precisely at the two ends: idea generation and commercialisation. Those are the points where human judgement, taste and appetite for risk matter most, the leap to a genuinely new idea, and the call on whether a market is ready for it. The review's own framing is that successful AI innovation depends on keeping a balance between technological capability and human needs, which is a polite way of saying the machine should not be handed the parts that require a human to mean something.

The practical read is a division of labour. Let AI multiply the middle so your people can spend their scarce creative and commercial judgement at the ends. It is the same reason you cannot out-hustle the machine: the value is not in doing more of the iterative work yourself, it is in the human calls that bracket it.

How do I put AI where it works in innovation?

By treating your innovation funnel as three distinct jobs and staffing each honestly.

1. Map the funnel: idea, develop, launch. Then look hard at where you bolted AI on evenly, and where it actually helped. The two rarely match.
2. Concentrate AI in development. Prototyping, simulation, testing, refinement. This is where the review finds it strongest, so this is where extra investment pays back fastest.
3. Keep humans owning the spark and the launch. Protect idea generation and go-to-market judgement as human work, and resource them as the scarce, valuable stages they are.
4. Fix the real blockers. The review is clear the constraint is rarely the technology: it is data silos, skills gaps and the struggle to scale beyond pilots. Clear those, or your development-stage gains stay trapped in a proof of concept.
5. Move now, because everyone is. New-product-development AI use nearly doubled in a year. The advantage in the development middle is being competed for this cycle, not next.

What does this change for me as a leader this quarter?

It turns "use more AI in innovation" into a sharper instruction: use it in the part of innovation where it is genuinely strong, and stop asking it to be creative on demand or decisive about markets. The compression AI offers in development, the ability to try ten prototypes in the time one used to take, is real and available now. The judgement at the ends is what you protect and grow, because that is where your people remain irreplaceable.

Do both and the shape of your innovation changes for the better: more shots on goal, taken faster, without surrendering the human judgement that decides which ones are worth taking. That is what the end of business as usual looks like in practice, not people replaced by machines, but a cleaner division of labour between what each does best.

About the author

Thomas Green

British technology futurist, AI keynote speaker and advisor. Thirty years across enterprise technology and AI strategy, helping leaders navigate the future of work. The futurist who died.