Project Learning Blog – Team 16

5/5 (1) As a member of Group 16, my group and I developed DIPS, the “Demand and Inventory Planning System” and designed it to integrate into LOAVIES’ workflow to streamline their inventory decisions. LOAVIES is a Dutch online fashion retailer with a strong online presence that introduces new styles every week. I was eager to work on the project as I wasn’t fully expecting the Information Strategy course to be this up to date with AI tools; I was pleasantly surprised.
My contribution to the team mainly revolved around the economic and ecosystem analysis. We decided to carry out individual research for each assigned topic and then coordinate the work to build the final deliverable. From a research standpoint, I was satisfied with the information I managed to gather for the economic analysis; on the other hand, I struggled with the ecosystem analysis. Indeed, since DIPS is not a platform, I had a hard time connecting this part with the course topics, as I felt, there is a heavy focus on platforms and network effects. Other than this I did not face substantial obstacles individually.
As a group we did have to iterate the project to make sure it was going in the right direction. In the beginning the idea was to take ownership of the whole Inventory planning value chain, and our main target was market-leading fast-fashion retailers. However, after consulting the IS Team by email, we realized that it would have been more tangible to take ownership of just one inventory decision for medium-sized fashion retailers. The original plan was to create an AI Agent to take ownership of the whole process. Indeed, DIPS’ original name was ADOE, Agentic Demand Orchestration Engine, as our idea was to gain a competitive edge over other inventory tools by automating the whole process of re-ordering and having no human involved in the process whatsoever. However, during our evaluation, we modelled an extreme scenario where from a strong external signal there is a 0% purchase conversion. We labelled this scenario “Full Hype”. During the Full Hype scenario DIPS would cause 36% more overstock compared to traditional forecasting methods. This helped us realize how important it is to have a human in the loop. External signals from social media can be easily misinterpreted and manipulated as there’s limited control over who can open an account and how many accounts are tied to a single person. After this realization we figured out that first, it wasn’t possible to rely exclusively on one social media platform, which is why we added Pinterest, TikTok and Instagram in addition to Google Search. Secondly, we realized that having a Planner Supervisor was also vital for our project to create value sustainably. We realized that the real value of AI augmentation doesn’t come from the increase in efficiency or the cut in costs alone, but rather from keeping human operators in the center of the loop and allowing them to make informed decisions with the best set of information possible.
One of the main surprises for me in this project was the scalability of information goods. Indeed, since we picked Track A we didn’t develop a traditional pricing strategy, we just computed the benefit that our implementation could bring to LOAVIES. I was uncertain whether the benefits would exceed the implementation costs. However, I was pleasantly surprised when I realized that for 30 items the payback period was 10 years. We picked 30 items as a benchmark since it’s the same number of items we had in the prototype and while checking the scalability we realized that by just doubling the items the payback period would fall from 10 to just 2 years. This is something I underestimated. I always considered Information Goods easy to scale since there is no physical production involved, although I considered this scalability a mere feature instead of a strategic advantage, because, in my opinion, if DIPS is a tool with an extensive economic potential, it is derived mostly by its scalability.
From a collaboration perspective I am satisfied with the engagement that my groupmates and I demonstrated throughout the project. The only thing I would have changed is the individual nature of the work. After we decided the direction our Inventory planner was taking, we split the work and we never checked in with other group members unless they explicitly requested. This can be considered efficient and time-saving, yet what I realized at the end of the project is that in the case where there is just one deliverable involved it is important to review each other’s work to understand findings, doubts, and projections since the final output needs to be assembled as one homogeneous product. This unfortunately happened close to the deadline and was our main pain point. Everyone’s work was solid, evidence-backed and connected to the course content, in our opinion. Although some parts of the project were heavily interdependent and because of our approach, they ended up quite disconnected. We deduced that the time we saved not checking in with each other was probably half of the time we spent fixing the flow on the final deliverable. For future projects I will propose weekly stand-up meetings to discuss any doubts, progress and future direction of the project. Furthermore, creating a shared workspace on an LLM could increase the coherence between the different parts even if the work is split between the group members.
My main takeaway from AI Strategy is that flashier doesn’t necessarily mean better. Nowadays, technological advancements are increasingly frequent, and firms are trying to catch up with these technological developments to keep their competitive advantage. This process became so natural and taken for granted. For instance, I believed that the best innovation could be attained through full automation and Multi-Agent AI Systems. What I would argue after this course is that the most value can be created from innovations that keep the human operator at the center of the strategic mindset and leverage AI and in general new technologies to help individuals make more informed decisions. Indeed, automation is feasible and convenient, but when the activity subject to automation is complex and not repetitive, human oversight can be vital. DIPS itself doesn’t aim to substitute LOAVIES’ employees, but rather to help employees overcome the limitations of bounded rationality through enhanced forecasting and GenAI. Furthermore, we managed to create the most value from a simpler tool than the original. Indeed, ADOE was broader and more technologically complex than DIPS; nonetheless, it performed worse in our evaluation. This is why the main takeaway that I have from the Information Strategy course is that digital disruption is not necessarily driven by the technology itself, but by how organizations use it to transform business models, processes and value creation.