A challenge I wanted to take on myself
During the Information Strategy course, we developed Smart Space: an AI-powered application that allows IKEA customers to put together complete room designs based on room measurements, budget and preferences. My contribution consisted of three parts: building the prototype, designing the user research and analysing the results.
Prototype development, in particular, immediately appealed to me. When I read the brief, I wondered how we could create a convincing application without any programming experience. Precisely because I’d never done this before, I wanted to find out. In hindsight, I learnt that a promising AI concept, a working prototype and a reliable end product are three different things.
From the first prototype to technical limitations
After researching suitable tools, I came across Figma Make. I spent a day learning the software and experimenting with various possibilities. Eventually, I had a first working version that I could show to my team. Our idea was thus no longer just a proposal on paper.
The biggest challenge began when we set out to improve that version. We didn’t just want to let users view three complete rooms; we also wanted to offer an interactive 3D room in which they could place, move and replace real IKEA furniture. This proved unfeasible within Figma Make. Consistently linking the correct images to the correct IKEA products also presented problems.
I kept trying to make adjustments and eventually spent a whole day sorting out errors. At a certain point, new prompts were barely yielding any improvement. That was frustrating, because I knew just how important visualisation and customisability were to our concept. Ultimately, I had to accept that we were testing a prototype, not a fully developed product.
This helped me understand a limitation that was also covered in the course material: a prototype can make an idea tangible, but it does not necessarily have to function like the intended end product. AI tools make the initial stages of development accessible, but for complex functions, technical expertise, reliable product data and additional systems remain essential.
Designing a fair comparison
My second responsibility was the research design. I didn’t just want to demonstrate that Smart Space worked quickly, but to investigate whether users actually experienced any benefits compared to the IKEA website.
That required more preparation than I had anticipated. I revisited the course material to identify which factors might skew our results. A key decision was to adopt a research design in which each participant used both systems. This allowed us to compare experiences within the same individual.
At the same time, someone might complete the second task more quickly due to prior experience of furnishing a room. We therefore designed four routes in which the order of the systems and the room types were varied. We also used fixed room dimensions and budgets to ensure the tasks remained comparable.
For the questionnaire, we chose Task Ease as the primary measure, supplemented by completion time, perceived control and choice confidence. I revised the open-ended questions several times, partly in response to changes in other parts of the project and feedback from team members.
I learnt that a careful experiment begins before you collect data. You must determine in advance what you want to measure, why, and what alternative explanations are possible. Even with counterbalancing, you cannot rule out all differences between participants, tasks and test conditions.
What the results actually showed
After the user tests, I analysed and processed the results. Sixteen participants used both systems. Smart Space was faster for fifteen of them: the median completion time was 2 minutes and 35 seconds, compared with 6 minutes and 12 seconds on the IKEA website.
This time saving did not surprise me. I had built Smart Space myself and knew that the system allowed users to put together complete rooms quickly. However, I was also aware of its limitations. The 3D visualisation was not what we had envisaged, product images were not always accurate, and rooms could not be adapted flexibly enough once they had been created.
The average Task Ease score was higher for Smart Space, but the difference was not statistically significant. Moreover, some participants felt they had little control. That, in particular, was an eye-opener for me: a faster process does not automatically equate to a better user experience.
It also became clear just how difficult it is evaluating the potential of a solution through a limited prototype. Participants react to what they actually see and can use, not to the application we ultimately want to build. At the same time, you shouldn’t simply dismiss negative feedback as a consequence of technical limitations. That feedback actually highlights which features are important for building trust and ease of use.
That is why I supported our conclusion to further improve Smart Space and retest it, rather than scaling up straight away. We still believe in the concept, but the research results do not yet justify a launch.
Collaboration and personal development
Collaboration within our team was pleasant and professional. We divided up responsibilities, agreed on clear deadlines and kept in constant contact. Whenever someone got stuck, others were willing to help find a solution.
The comments in our shared document were useful. They enabled us to question each other’s choices, discuss ambiguities and ensure that different components aligned better with one another. I discovered that being independently responsible for a component does not mean you have to develop it in isolation. It was precisely by regularly exchanging feedback that our end result became more cohesive.
I learnt to take the initiative when faced with unfamiliar tasks. Despite having no previous experience with Figma Make, I managed to develop a usable prototype. At the same time, I had to learn to accept that not every technical ambition could be realised within the time available.
A different perspective on AI-strategy
For this project, I saw AI primarily as a way of making existing processes faster and simpler. Smart Space confirmed that potential, but also showed me that speed alone is not enough.
An AI application must use reliable information, provide understandable recommendations and give customers sufficient control. For Smart Space, this means, amongst other things, accurate IKEA product data, a credible room visualisation and the ability to rearrange and replace furniture.
I now also have a better understanding of how valuable data can underpin a strategic advantage. With reliable product information and user feedback, Smart Space could improve its recommendations. However, a positive feedback loop is not a given: first, it must be demonstrated that users trust the system and continue to use it.
If I were to start again, I would first investigate which features are technically feasible and essential for the test. This would help me prioritise and identify technical dependencies earlier.
My key lesson is that AI strategy isn’t about automating as much as possible, but about demonstrable customer value. Having built, researched and evaluated Smart Space myself, I now have a better understanding of just how much work lies between an attractive idea and a reliable application.
