Student’s role and contribution
With the aim of maximizing our chances for success in the AI strategy Lab, we split the work evenly among the 4 group members, with each one working autonomously on the respective area. In the process we made sure that all the parts of our work align, so that no confusion is generated. My role was to work on the Economics part, and also Ecosystem and Defensibility. For the Economics Model, it was very important to build a base, in form of assumptions, for the further calculations. I assumed a certain amount of returned items to be flagged by the AI model, a certain amount of human labor to be involved in the updated return process, the amount of time that it takes for a store employee to check a return item, and so on. The economic model was built in relation to the first million of items purchased, assuming a 25% return rate. The costs are calculated for building the AI model setup, as well as for maintaining it, and also the costs for the reduced human labor are taken into account. As there are Fixed yearly AI costs in the model, economies of scale are assumed for a growing amount of users. Not to forget to mention the Scalability, Pricing & Revenue Logic, and the Expected Benefits subsections. For the Ecosystem and Defensibility part I had to determine who are the Dependencies and Complementors of the model, who could pose an Envelopment Threat to the model, and whether is there any possibility for the Model to defend itself.
Important decisions and iterations
The most important decision that we took, and that ended up being crucial for the feasibility of our project, was moving from the idea of having a picture-taking conveyor (and also involving a significant amount of human labor in the process), to a concept where customers take pictures of their returns and upload them to a specific section in the Zara app, where our AI model is present. Another vital moment for us was the generation of the prototype and conducting multiple tests to see how it copes with its functions. Moving from the group perspective to the personal one, I should mention that in the process of conducting the economic model I had to make important assumptions, some of them being risky, to make sure that there is a base on which the economic model can be built. For example, I had to assume that only 20% of the returned items will be flagged by the AI as possible fraud. It might be a realistic assumption in the long time, but in the beginning it is likely that this number will be way bigger, making the project economically unfeasible. To continue, another assumption was that a share of the successfully AI processed returns, amounting to 15% of the total returned items, will be checked by the humans to assess how accurately is the AI model doing its job. This number is also likely to be reached only in the future, when the AI model is improved based on the received feedback. Ultimately, it leads to more human labor necessary than assumed, which has economic consequences for the model.
What surprised or failed
What surprised me was the fact that with AI as an Information Good you cannot benefit from economies of scale as much as with other Information Goods. Namely that each AI activity, how little wouldn’t it be, costs additional money to implement. What failed was the weakest assumption. Although initially having assumed a flag rate of 20% by the AI, the prototype sent 50% of the items for further check by a human, making the project economically unfeasible under such conditions.
How feedback and evidence changed the work
Within our group, we had interesting debates about how should our project look like, we changed the concept a few times, shaped it so that the final idea makes sense for our target company (Zara), but also that it ends up being economically feasible. At the end we arrived at the idea, as mentioned above, that the customers should send pictures of their returned items to the designated AI return model section in the Zara app. We were even going through a moment when we were thinking about making our project more capital intensive, but that capital intensity would end up making it also more labor intensive, making the final project economically unfeasible. But some time before that we gathered for our first call, and had to decide what will we be doing at all. We were four people with four completely different ideas. For example, my initial idea was to create an AI model that would be supervising the banking transactions, and assessing whether any of them can represent fraud or scam activities. That would have been done by promptly sending a report to the bank. Ultimately, we consulted the professor and she suggested that our project should represent something realizable and down to earth. That’s how the professor’s feedback put the basis of our work, and we ended up focusing on the idea of my friend Michal.
What did I learn from collaboration
I have to mention that it was an enormous pleasure to collaborate with the members of my group. We always maintained a very friendly and helpful atmosphere. As a colleague mentioned after the submission, it was truly a team effort. From this particular collaboration I learned that in a group project, if each member has different ideas, sees things differently, it is essential to make compromises in order to arrive at a solution that suits everyone. I learned that when people with different backgrounds gather together for a project, knowledge sharing is the key to succeed. I also learned that when every member is working on a different part of the project, it is mandatory to have constant contact so that everything stays aligned, and everybody follows the same path.
How the project changed my understanding of AI and business strategy
As already mentioned, I contributed to the group project by working on the Ecosystem and Defensibility, but most importantly on its Economic model. By working on the Economic model I, understandably, estimated the relevant AI costs. Both from the lecture slides and from working on this project it was a novelty to witness how, compared to usual Information Goods, AI generates costs for every single use, as in our case we have to buy tokens for every single analysis of a returned item. It was important to include this in the pricing model, ultimately determining that we have 3 types of AI costs in such a project, one-off costs, fixed yearly costs, and also variable costs per returned item. Overall, by working on this project, generating ideas, and also witnessing the work of my colleagues, I realized how much more AI is able to do nowadays rather than just generate answers for some study questions. I also realized how much easier it is do business with AI, if the model is so efficient that it reduces the human labor factor in your business strategy.
