From an empty flat to an AI strategy
The boxes were unpacked, the walls were bare, and I had no idea where to start. Most of our team had experienced that problem, so it became the starting point for IKEA RoomMate. It was interesting how the theory from the lectures translated to the real world: every good idea starts from the customer problem, not the technology. The key question was: what bothers us and why? Our first challenge was whether the solution should be sold to all furniture stores or be an in-house project for one of them. We chose IKEA as it already has Kreativ, a good base to build upon. RoomMate is a GenAI layer on IKEA Kreativ that combines a room scan and a short prompt (style, colours, budget) into a full room design with real, buyable IKEA products.
My part: problem, process, and keeping us on track
After having a draft of the problem and solution, we divided the tasks of this assignment. I covered the problem and opportunity section as well as the governance and failure analysis. Additionally, I drew a BPMN 2.0 process map to serve as the basis for the entire assignment. This made sure we all worked on the same solution and did not drift apart. Because I also worked for a year in consulting between my Bachelor’s and Master’s, I took over part of the project management and slide design.
Too many choices, too little help
Why did we all feel decorating a room or flat was such a burden? Part of the answer was choice overload and buying fatigue (Iyengar & Lepper jam study). Papers and the lectures also showed that online shopping worsens this problem because thousands of products are listed with no extra costs, which delays buying. This identified the problem, but the solution needed further research. IKEA Kreativ already scans rooms and shows 3D furniture, but it does not help with choosing or designing. I identified this as the customer friction and our opportunity. To validate it, I applied lecture concepts: Porter’s Five Forces, Clemons’s “newly vulnerable” test, and Barney and Reeves’s question of which scarce resources AI strengthens and which advantages it commoditises. I concluded that IKEA’s moat is not GenAI itself, but AI combined with real products, stock and delivery.
Process map
I learned BPMN 2.0 during an internship. It shows processes in a structured and logical way. Firstly, I identified the swimlanes and basic flow, then added loops where needed. Team feedback meant answering things like how many feedback loops do we need, when a human should step in, and what IKEA departments do we need to create. We went through the process many times, adding in the legal requirements and oversight by humans, in particular.” We decided in the end that we only need one new unit and we’ll put humans in after the AI fails three times. GenAI took those decisions and hand sketches and made a clean diagram out of it. My biggest takeaway was that you need constant feedback and collaboration for an in-depth, complete process. It was also fascinating how well AI helps once it gets the logic and drawings from us.
Designing for trust and compliance
The legal topic was totally new to me. Sessions 6 and 8 really helped me understand GDPR and the AI Act. RoomMate is not high-risk under the AI Act, so human oversight was not legally required, but we decided to include it anyway. Session 7 taught us why, because RoomMate adds seller uncertainty. Customers may ask whether the AI is choosing for them or for IKEA. That is a classic principal-agent problem, and once customers trust the tool, IKEA could quietly push bigger baskets (moral hazard). So we designed against it. Every pick shows why it was chosen (signalling). Every item comes with a cheaper alternative so customers can compare (screening), and the budget is a hard limit. We also measured success by the share of the suggested list that customers buy, with a check on basket size against budget, following Kohavi and Thomke’s idea of one overall evaluation criterion plus guardrails. From then on, governance wasn’t just compliance. It became part of why customers would use RoomMate at all.
Teamwork
My favourite part of this assignment was the teamwork. We thrived due to our differences. During our bachelor’s, we studied Computer Science, International Business, and Information Science. With a Computer Science background, one student could judge what was technically realistic. With my International Business background and consulting experience, I looked at the process, the costs and the organisation. And after learning Figma during the Information Science bachelor’s, another teammate built the prototype from the customer’s view. We constantly questioned each other’s ideas, assumptions, and even wording, which helped us improve the project. For example, we discussed VR, where we decided against relying on hardware because it would create additional barriers, especially because our target group has limited time and money. This taught me that the most advanced technology is not always the best solution for the customer.
Challenges
Our research showed recommenders raise conversion by only about 7.5%, and none proved that customers buy most of a suggested list, so we weakened our core assumption to “a substantial share”. When working through the whole document again, I found errors in the business case. This was stressful but very valuable. At work, inputs came from the client, but in this case all financial assumptions had to be developed and justified by ourselves. The rebuild also changed the story. The sensitivity analysis showed that at a 1.0 percentage-point conversion uplift, RoomMate never breaks even. At 1.5 points, it pays back in month 19. Our honest recommendation therefore wasn’t “launch it” but “test it”: a three-month A/B pilot of RoomMate against Kreativ, scaling only if the uplift reaches at least 1.2 points and the guardrails hold. Session 6’s point that strong opinions can’t replace a control group suddenly mattered a lot.
How did my view of AI strategy change?
Before taking the Information Strategy course, I thought of AI strategy as adding an AI feature. Afterwards, I understood it as asking what AI makes abundant and what stays scarce. Good room designs will be cheap and everywhere. A customer’s personal AI agent will be able to produce one too. What stays scarce for our IKEA RoomMate idea is product data, stock, delivery and trust. That’s why our future-market test ended where it did: a generic agent can suggest a room, but only IKEA can deliver it to your door.
