Designing AI-Based Decision Support for Sustainable Consumption: A Prototype-Driven Study in Digital Retail Services

Sammanfattning

Many consumers say they want to shop more sustainably. But most retail websites do not make that easy. Walk through a typical online furniture store, and you will find sustainability information buried in product specifications or tucked away behind a separate filter tab. This happens right when you are already trying to balance price, looks, and whether the thing will actually fit in your living room. This thesis asks whether an AI-based decision support system, built directly into a furniture retail website, can change that. We built a working prototype called IDEA, which stands for Interactive Decision-support Environment for Sustainable Action. It has three main parts. First, a conversational AI powered by the Claude API. Second, an eco-score for each product from zero to ten, shown alongside the estimated CO₂ savings. Third, four product categories shown together on the same page: brand new, second-hand, refurbished, and eco-certified new. The study is qualitative and interpretivist, following a Design Science Research approach. We collected data from 204 backend interaction sessions split between an AI-assisted condition and a control condition. We also ran a post-use survey with 106 participants, conducted ten semi-structured consumer interviews, and talked to one IKEA digital professional as an expert interview. The interview data and open-ended survey responses were analysed using Braun and Clarke's (2006) thematic analysis framework. Within this sample, the AI's main job turned out to be making things visible, not persuading people. When second-hand and refurbished products appeared alongside new ones by default, and when a single conversation question surfaced those options, the AI-assisted condition produced an 87.9% sustainable item selection rate. In the control condition, with the same products and prices but no AI support, the rate was 40.4%. Trust in the eco-scores was moderate, around 60 to 70%, and participants said they would trust the numbers more if an independent organisation verified them. People also reported that they did not feel pushed around. The design preserved their sense of autonomy, which matches the libertarian paternalism idea that a nudge should steer without constraining. On the question of deployment, the study finds a real opportunity that is technically within reach. IKEA already has the data infrastructure and the in-store interaction model to support something like IDEA. But there are five governance challenges that would need to be addressed before responsible deployment. First, a commercial conflict of interest: the retailer profits from new products, not second-hand ones. Second, technical integration: second-hand inventory sits in a separate system. Third, regulatory immaturity: there is no standardised way to certify eco-scores across markets. Fourth, cultural bias: the European lifecycle assessment framework does not fit how some users think about sustainability, especially durability and repairability. Fifth, the carbon cost of the AI itself: running conversational AI at retail scale has its own environmental footprint, and that has not been accounted for. The thesis makes three contributions. Theoretically, it brings together practice theory, nudge and behavioural theory, service-dominant logic, and human-AI interaction theory in a way that has not been done before in this context. Methodologically, it shows that a prototype-driven Design Science Research approach can generate real interaction data, not just responses to hypothetical scenarios. Empirically, it adds to the small but growing body of research on how users actually experience a live conversational AI sustainability assistant in a furniture retail setting.