AI can reinforce unrealistic product design expectations

AI-generated summary

AI-generated product concepts are changing the enquiries received by product design consultancy Npd, but the author argues that convincing visualisations can reinforce unrealistic expectations about product viability, development costs and the work needed to reach manufacture. Drawing on 20 years in the industry, the author describes a growing pattern among start-ups: prospective customers arrive with products they consider already “designed” using AI and expect the consultancy simply to make them real. This shift has both financial and practical consequences. In one discussion, the author estimated that the conceptual work a prospective client presented would previously have represented around £20,000 in consultancy fees. AI allows start-ups to save that initial expenditure and bootstrap ideas, while potentially increasing the number of enquiries reaching design agencies. However, the author argues that professionally led concept development would likely produce a better product and simplify subsequent development. Bringing specialists in only after a concept has been established also limits their ability to shape the fundamentals—an important concern for design for manufacture. AI-generated cost estimates can further distort expectations, whereas, in the author’s experience, early specialist input often avoids greater spending on tooling, redesign and getting to market. To assess the change, Npd collected several months of information about inbound enquiries, including how customers found the company, why projects were rejected and the likely value of work associated with AI-generated briefs. Estimated average project value was approximately £15,000, compared with the consultancy’s £50,000 target. The rejection rate rose to 89%. The company rejected almost all these enquiries or discouraged customers from proceeding because it considered the ideas weak or inconsistent with its ethos. These figures describe Npd’s own experience rather than an industry-wide study, but they underpin the author’s concern that increased enquiry volume is not translating into suitable, qualified work. The central criticism is that AI can visualise an idea without adequately challenging its premise. The author characterises current frontier models as endlessly accommodating: they respond to requests without the fatigue or practical concerns that shape a human designer’s work. This responsiveness can mislead founders when it is not accompanied by scrutiny of market suitability, design details or user experience. In the author’s view, an attractive representation does not validate demand, and AI cannot turn an inherently poor idea into a product customers will want to buy. The article also identifies a contrasting pattern among experienced product-development organisations. Npd’s regular clients are not turning to AI in the same way, and proposals from established companies are not being formulated by agents. Several senior design managers consulted by the author said their companies had not permitted them to introduce AI, most often because of concerns about intellectual-property ownership and compliance with cybersecurity legislation. Based on these conversations, the author expects adoption in those organisations to be slow and limited. Npd’s core development process remains largely unchanged. It continues to use Onshape as its preferred CAD software and Keyshot for visualisation, while solving problems through whiteboard work, discussion and observation involving experienced people. The consultancy does seek to automate administrative activities such as project planning, note-taking, data analysis and reporting, but its client-facing concept work remains human-created. The author concludes that designers and engineers should use AI to improve efficiency rather than let it lead product development. User experience, ergonomics, sensory impressions and emotional responses to new products are presented as areas current frontier models do not understand, and the author doubts computers can replicate these human dimensions. That longer-term judgement is explicitly qualified by an acknowledgement that rapid AI progress could make the article’s conclusions age poorly.