How it started
In the beginning of the project, when we were brainstorming about the domain, one teammate who works in healthcare pointed out the problems with administrative workflow in the industry, which later became the field we landed on. However, identifying and specifying the problem was not easy, as the healthcare industry is broad with multiple complex workflows in each submarket. Due to my little personal knowledge in the healthcare industry in the Netherlands specifically, it was difficult and even a bit confusing what to tackle in the beginning. Yet luckily, one teammate who has technical background and domain knowledge of the industry, guided the rest of us through the process and came up with the solution in a collective effort. Considering that it is not feasible to create a whole paradigm shift, we decided to narrow it down to the ‘DBC optimization’ process within the medical rehabilitation market.
CareFlow
Careflow is an AI agent that analyses Diagnose Behandel Combinatie products (DBCs) for medical professionals in rehabilitation centers. It automates the time-consuming manual check of patient treatment hours against revenue and thus enables the workforce to have more time on higher-value and patient-related tasks. By increasing accuracy and efficiency, the value was created.
My role and contribution
We divided the tasks according to our individual strengths, and I took responsibility for the economics, pricing model, and the value part of the business model. I approached the analysis with the following questions: what value do we create and where to capture value from? What is the cost structure and which pricing model can we adopt?
I started with confidence based on the nature of information goods, where the initial development costs are significantly high and the recurring marginal costs later are close to zero. This has shown to be not completely correct in the model. In the system, the marginal cost per DBC analyzed is near zero, but the recurring human costs on acquiring a new rehab center/department remain substantial until onboarding is standardized. This limits the following scalability too. Regarding the pricing model, we started from charging a usage fee, after each DBC case is analyzed. However, the more I reflected on the structure the more doubts I had in mind. The main concern was the potential discouraged usage especially from small and more cost conscious organizations, which put them in a cost versus process improvement dilemma. Thus, later on, I came up with a three-tier versioning model, where the annual subscription fee is charged and plans differ mainly in DBC allowance credits. This encourages usage as organizations self-select the plans and the fixed annual fee is also easier billed.
The results
The results surprised me positively: the AI Agent is able to reduce the time spent by more than 80%. This gives us a strong foundation to build on. The positive response from the medical professional with excitement about the official launching of the system, also shows qualitative evidence to validate initial demand. However, due to time and knowledge constraints, we were not able to test the accuracy target of the system. Considering the complexity nature of medical measurement, which requires closer look into a patient’s treatment plan and schedule to assess, we are only able to track the schedule for assessment. This leads to the prediction uncertainty, which should be tested further with medical professionals.
Collaboration
Collaborating with the team taught me how valuable the diversity within a team is. We came up with ideas from different perspectives, requiring us to constantly align our outputs but at the same time allowing us to approach the problem differently. The exchange of ideas broadened my understanding of the project domain and . The diversity in experience also built a foundation to help out each other, especially the technical expertise from one teammate. Learning from each other’s strength not only facilitates collaboration but also amplifies the effectiveness and efficiency of solutions.
Take away
Before this project, I viewed AI simply as a tool that assisted me in learning and answering my daily questions. I also underestimated the possibility of how much value an AI can create by saving time, increasing the efficiency and task accuracy. The project not only broadened my knowledge in Agentic AI but also more importantly, how we can utilize the agent to solve the relevant problems. I also learned how AI can optimize workflow to reduce human burden. I also found out that it should be used as a tool to ease more time for humans, rather than simply a replacement. This resonates with the importance of keeping the human in the loop to reinforce the guardrail.
Seeing and learning about the projects of other groups broadened my perspectives on problem identification. I realized that finding a problem to solve does not have to be too complicated or tackling a large-scale issue straightaway. Conversely, it could be found everywhere, even just something small that emerges from our daily errands. With the critical lens developed through the project, I will approach future projects by looking closely at daily challenges, reaching out to actual customers to fully test and validate the use, ensuring that they address the real needs and thus offer opportunities for meaningful improvement.
