Innovation with AI: From Ideas to Experiments
Move beyond generating more ideas with AI. Design experiments that challenge customer assumptions, reduce uncertainty and support investment decisions.
You ask AI for new product ideas during a meeting. Within minutes, a long list appears. The team likes a few, a presentation follows, yet the customer need remains unverified. Faster idea generation does not necessarily reduce innovation uncertainty at the same speed.
In How to Drive Innovation with Generative AI, Alan Iny, Luc de Brabandère and Justin Manly discuss using generative AI to question strategic assumptions alongside generating ideas. This broadens the question an innovation effort starts with. Rotman Management, ROT507.
Change the frame before increasing the volume
“How can we improve our product?” puts the existing product at the center. “What is the customer waiting to accomplish?” begins a different investigation. AI can help vary these frames, but deciding which is meaningful still requires real customer context.
A customer complaining about slow service, for example, may primarily need to know when the result will arrive. Speed, visibility and reassurance represent different solution spaces. Without separating them, a hundred ideas may simply be variations of one assumption.
Put a learning question beside every idea
One working practice we suggest is to turn the idea list into a small experiment table:
Idea | Critical assumption | Smallest learning step |
|---|---|---|
Live status notifications | Uncertainty matters more than waiting | Compare two notification examples with customers |
Personalized recommendations | Customers value personalization | Deliver a manually prepared response to a real need |
Automated initial assessment | Greater speed does not reduce trust | Observe responses and perceptions in a controlled group |
These are illustrative examples, not outcomes from a client engagement. Their purpose is to clarify what must be learned before an expensive solution is built. A useful experiment produces information that can change an investment decision, including a decision to stop.
Separate synthetic feedback from customer evidence
An AI-generated response in a customer role can be a useful hypothesis during preparation. It does not replace an actual customer’s behavior, willingness to pay or product experience. Preparing interview questions with AI is different from establishing what customers want.
At the end of an experiment, ask which assumption strengthened, which weakened and what should happen next. These questions help an innovation portfolio develop beyond attractive presentations.
Our AI-Empowered Leadership program develops the skills to challenge assumptions, redesign team workflows and turn an agent prototype into a measurable experiment. Start the next idea meeting with a customer question that needs an answer, and use AI to improve the range and clarity of the possibilities you test.
