Team role and contribution
In order to come up with an idea for our project, I proposed that each of the four members prepare a mini case before our first meeting. This is how we came up with CareClear: an application that uses AI to translate and explain medical letters in the patient’s own language and then uses the information to create personalized scheduling for medication and appointments. Regarding the content contribution for the assignment, I was responsible for the prototype, the value creation and pricing of the solution, the AI canvas and the formulation of the causal test. Especially the creation of the prototype was exciting for me. In the team I took up a less visible role, one in which I kept the team aligned through group discussions and one-on-one calls. At times, I took a more prominent role in guiding the project when I felt that we were losing track. I believe my contribution helped the team’s communication and our understanding of CareClear and who it was for.
Key decisions and iterations
The initial idea for CareClear was to create an aftercare ecosystem and build an application with all the possible features that we could think of: translation, scheduling, insurance, a chatbot, GPS to pharmacy locations and more. After receiving feedback on our initial proposal, we were advised to focus on one of the proposed features (translation) due to the time limits of the assignment. Looking back, defining the initial idea as an “ecosystem” was incorrect, what we were aiming for was an orchestrating role between hospitals, pharmacies and insurers. By narrowing our idea to only implement translation and scheduling, our role moved to that of a complementor.
Surprises and failures
What I believe to be our biggest failure is customer identification. Throughout the journey, we were only considering the patient users and the hospitals as possible customers. Near the end of our assignment, I realized that the costs of additional GP and hospital visits are covered by the insurance companies, not the hospitals. Although the focus on the social value made a strong case for the willingness to pay for a hospital, I believe concentrating on the insurers as customers would complement the economics of the solution better by focusing on the prevention of additional medical visits and the costs that come with that.
A big surprise for me was that there was no dedicated product for the translation and explanation of medical instructions. After some research, I understood that the policy context is less supportive of our idea than I initially expected. In 2012 there was a cut in the Dutch interpreter funding, the reasoning was that patients should be responsible for overcoming the language barrier (Broersen, 2014). This made me question whether our idea is filling a gap that was deliberately created.
Another setback came up during the creation of the prototype. For this part of the project, I chose Lovable as the platform to work with. The creation was running smoothly until I ran out of credits, which resulted in me having to transfer the project to Claude via GitHub. This taught me about the risks of relying on external platforms. This dependency also occurs in our project, as CareClear is reliant on the Electronic Health Record (EHR) vendors and an external LLM.
Feedback and evidence
The causal test we conducted revealed problems with the prototype. Even though the technical capabilities of the prototype do not count towards the assignment grading, there were still several issues found after prototype testing. The system translated Dutch medicine names to other languages directly, which was not always correct as brand names differ globally. The translations still used a lot of medical jargon in explanations, something we wanted to prevent to make the instructions more accessible. Finally, the timing of medicine intake could be interpreted differently in different languages; one interpreted “as needed” as a morning intake, while another language advised a morning and evening intake. All these issues were fixed by preserving medicine names as stated in the letters, implementing plain everyday wording and introducing a “double check” warning when intake was not indicated clearly instead of guessing.
What I learned from collaboration
Within our team, the work was divided fairly, and the allocation was based on our personal preferences. The overall collaboration ran smoothly with frequent meetings and internal deadlines. One problem we encountered throughout the project is that having each member be responsible for separate parts of the project can create minor misalignments that lead to bigger problems if they are not identified. Throughout the project, CareClear went through numerous changes. While comparing our parts, we discovered that many minor changes were introduced by each of us, resulting in a lot of inconsistencies. When I noticed these inconsistencies, I proposed having a meeting and discussing the final version of the idea. During the call, I documented the details we discussed to have a visible overview. For future assignments, this issue can be prevented by having the idea written out in a shared document, which can be adjusted when members implement changes. In addition, I should have asked for feedback from my teammates more actively. Overall, I have learned a lot from collaborating with the amazing people from Team 5. Their creativity, handy tricks, clear communication, and perfectionism have given me a good perspective for my personal improvement.
My changing view of AI and strategy
Before the start of this course and the assignment, I believed that the implementation of the technology behind the AI was enough to create value. Looking back, I see that the value of CareClear lies in the AI being used to complement valuable data from the EHR to match the demands of the users. This was further illustrated by the findings of our causal test in which we tested whether CareClear would be a better tool for understanding medical instructions than Google Translate. The result showed that both the group using CareClear and the group using Google Translate were able to identify the prescribed medicine correctly. However, on questions regarding frequency, timing and duration, CareClear scored 92% compared to Google Translates 68%. The participants also mentioned that for them the value lay not solely in the translation but mainly in the personalized medicine scheduling and appointment dashboard that was automatically updated for them.
This made me realize that AI implementation is only a small part of the strategy. A company will not capture the value of AI if competitors can copy your technology. An EHR vendor such as ChipSoft can introduce translated summaries to the data it already holds at a low cost, as the variable cost of AI is only a few cents per patient. This shows that power lies with whoever holds the data. This aligns with Yerramilli-Rao et al. (2025), who argue that when AI makes expertise cheap, the value moves to whatever remains scarce.
References
Broersen, S. (2014, March 20). Artsen in de knoop zonder tolkenvergoeding [Doctors in a bind without interpreter reimbursement]. Medisch Contact, 582–583.
Yerramilli-Rao, B., Corwin, J., Li, Y., & Lakhani, K. R. (2025). Strategy in an era of abundant expertise. Harvard Business Review, 103(2)
