AI can make an unresolved idea look like a finished product. A proposition that would once have stayed a paragraph can become a pack, a demonstration or a scene from someone’s day within an afternoon.

The important question in AI-assisted concept development is which choices the picture has already made on the team’s behalf, before a gate review, a retailer or a consumer in a test responds to it.

The picture contains a proposition

Imagine an early idea for a refillable personal-care range. It’s an illustration, though a familiar kind of development problem.

An AI-generated image shows a beautiful bottle on a bathroom shelf. The description beside it promises convenient refills and less packaging.

The image makes the idea tangible. It also suggests a material, a size, a place in the home and a level of quality, and the words suggest that refilling is easy. One reviewer pictures inexpensive pouches delivered when needed; another assumes a subscription. Nobody has agreed either arrangement.

A positive response would still leave open what attracted the person, whether the appearance, the environmental ambition, the convenience they imagined or the whole combination, and whether the business could deliver that combination at a price they would accept.

None of those choices needs to be resolved before research begins. Early research exists partly to resolve them. They do need to stay visible as choices, so they don’t become facts because a persuasive image includes them.

Use the speed to explore the choices

Faster visualization is useful here. The team can make different versions of the opportunity concrete enough to discuss before investing heavily in one expression.

For the refillable range, a durable object people want to keep supports one proposition. An inexpensive system that asks almost nothing of the user supports another. Seeing both reveals a strategic choice that a single polished image conceals.

Those alternatives need enough definition to be understood, not every incidental detail settled. The purpose is to learn which differences matter, including differences the team had treated as secondary. Extra capacity is best spent on an alternative the team would otherwise leave unexplored, and more output doesn’t by itself provide new evidence about customers.

Polish changes what a score means

A finished demonstration and a thin paragraph are different invitations to imagine. When a polished option is tested beside a plain one, or an AI render is scored against norms built on simpler concept boards, the result can’t be read at face value: part of the score is the finish.

Testing options at the same level of finish, and against benchmarks built the same way, keeps the comparison about the idea. This problem predates AI. Generative tools make it easier to meet early, when a plausible expression can arrive before the decisions behind it.

A convincing first version can become an anchor

In a CHI 2024 experiment, people sketching ideas with an AI image generator fixated more on an initial example than those with no support. It was one design task and one tool, but it is a reason to watch how early pictures shape what a team explores next.

The person anchored by a render is often not the consumer. It is the executive who sees it in a gate paper, or the retailer who sees it in a sell-in deck, and who then remembers the idea as finished.

A format, occasion or audience can start to feel part of the concept when it was one possibility among several.

AI can loosen that attachment too, by making another direction easier to imagine. Its contribution depends on whether the team uses it to explore a choice or to keep embellishing the first answer.

The right degree of finish depends on what you need to learn

Some questions require realism. A service can be hard to judge without experiencing its sequence, and a packaging idea can depend on how it appears among alternatives. Removing the detail would remove the very thing the research needs to understand.

The same goes for working prototypes. A feature demonstration built in a day can imply that something can be made, at a speed and a cost no one has yet proven. The aim is to make clear what the stimulus demonstrates, what it invites someone to imagine and what the team is still considering.

An encouraging reaction might justify developing the refill experience further, or exploring a different price or format. It doesn’t settle the whole proposition.

AI gives innovation teams a valuable way to bring possibilities into view earlier. Before a team backs one, it should be able to say which of the choices in the picture the evidence supports, and which it simply inherited. An idea can look ready long before the business knows what it is ready to back.

Common questions

Do AI-generated concept images change concept test results?

They can. Finish changes how people respond, so a polished AI render tested beside a plain concept, or scored against norms built on simpler boards, can’t be read at face value: part of the score is the finish. Testing options at the same level of finish, against benchmarks built the same way, keeps the comparison about the idea.

How should innovation teams use AI in concept development?

Use the speed to make different versions of the opportunity concrete enough to compare, and keep the choices a picture implies, such as price, format and service model, visible as choices until evidence settles them.

Sources

  1. Wadinambiarachchi, S., Kelly, R. M., Pareek, S., Zhou, Q. and Velloso, E., “The Effects of Generative AI on Design Fixation and Divergent Thinking,” Proceedings of the CHI Conference on Human Factors in Computing Systems (CHI ’24) (2024). Peer-reviewed conference paper; a between-participants experiment with 60 people on one visual ideation task, comparing an AI image generator, image search and no support.