How does one disrupt administrative workflows in the healthcare industry?

The question was the start of the project and I can tell you; it is not an easy question to answer. The importance of the question is increasing monthly as in the healthcare industry is facing the imbalance between treating more people with less healthcare professionals.

In this blog post, I will take you with me on the journey of creating Careflow from my point of view; what my role and most important contributions were, what I learned from each step and how my understanding of AI and business strategy changed.

My main role in the project was to investigate, collect and create content for the project and create a prototype.

Careflow
Careflow is an AI agent that reduces the administrative burden of optimising Diagnose Behandel Combinatie products (DBCs) by 83% for healthcare professionals in the rehabilitation market. DBCs are a container where treatment hours are collected in a period to reimburse a medical product to the insurers. Each medical product has a hour range with each a price than the previous hour range. The more optimised the DBC is, the higher the net return based on the revenue of the reimbursed product and costs of treatment hours spent.

Investigation
The healthcare industry is historically part of the late majority and laggards in terms of adopting technology, which poses its additional challenges when advancing from creating prototypes to selling products. Selling products to the late majority and laggards require different business strategies than selling to early adopters and the early majority. These adopters are notorious for only adopting technologies when they are beyond proven or simply when they are forced to adopt to stay in business.

Furthermore, The healthcare industry is broad and each submarket of the healthecare industry has numerous workflows. I learned that it is not realistic to create a paradigm shift within the healthcare industry with a project that lasts 6 weeks. That shifted the goal to narrow down the scope of the project to a workflow in a submarket of the healthcare industry.

I identified a couple challenges from the investigation:
which part of the healthcare industry do I want to focus on and which workflow do I want to improve?

To tackle the first challenge we researched that the healthcare industry. The healthcare industry is divided in three levels of healthcare. The first level provides general care, the second provides specialised care and the third provides highly specialised and complex care where first and second levels are not sufficient to treat patients.

Furthermore, in the third level medical rehabiliation care, most patients receive long-term care and that there are many administrative workflows that can still be automated by using proven technologies (e.g. Agumented Reality (AR) or GenerativeAI (genAI)) and emerging technologies (conversational AI (CAI) and agentic AI). Furthermore, the medical rehabilitation market counts 34 organisations with a shared revenue of €825 million in 2023 according to the most recent market report.

You would think: Great, the first challenge is done! However, the second challenge posed a different threat. The medical rehabilitation market has many specialised workflows that you can improve.

Where do you start when there are many workflows?
I interviewed a program manager, project leader, information security officer and doctors from a medical rehabilitation center. Their ideas ranged from reducing the administrative burden on analysing referals, communication traffic between colleagues, creating prescription letters, the administration of patients’ examination scores, no-shows, appointment registration and analysis on optimising DBCs. It surprised me that they had a varying number of ideas that shared overarching topics: communication traffic and optimising DBCs.

An important decision was made to pick the workflow ‘optimising DBCs‘ over ‘communication traffic’. While ‘communication traffic’ sounds easier on paper, it is not. The candidate workflows for ‘communication traffic’ require extensive medical knowledge to create a more efficient and reliable workflow. Furthermore, testing the workflow will require experts. For the scope of the project, this is not realistic to achieve. The workflow ‘optimising DBCs’ requires manual analysis of multiple platforms to create a recommendation that a physician approves or rejects.

Prototyping
Creating the first prototype taught me a life lesson: the best way to gain hands-on experience with emerging technologies is by prototyping!

Collaborating taught me that a prototype should not be only ‘technology-push’ by a single person, but that it also requires proper testing, input and incorporating the gathered feedback from the whole group to create a prototype that all of us understand. For example, after creating version 30 of the prototype, the prototype had the information spread out across the page. What I perceived as tacit knowledge and an easily readible spread of information, was not perceived as tacit knowledge and understandable. One could say the prototype is a fail at this point. However, the gathered feedback led to a more intuitive prototype with a tutorial that is easy to understand. This feedback loop was kept until we finished version 87.

AI Strategy
At the start of the project, genAI was picked as the technology for Careflow. However, the guest lecture by Unframe explained agentic AI thoroughly and provided evidence that agentic AI is a better fit for the target workflow. Looking back, the iteration was the right choice as genAI requires human input to create output and agentic AI is a continous autonomous loop from event trigger up and until recommendations. This taught me that understanding creating an AI strategy starts with understanding the alternatives and differences.

Furthermore, the project taught me that the there is more than just the ‘AI’ in the strategy that is often overlooked: the governance of an AI technology. It is not only about creating the prototype, testing and validating, but also what is the logic behind the AI technology and its workflow, which privacy laws and (international) regulations does it need to pass in order to be compliant and which risks does using the chosen AI technology pose?

Business strategy
Throughout the multiple prototyping one lesson learned kept bugging me: How do we adapt the business strategy in the healthcare industry for the late-majority and laggards?

Collaborating with my group helped me understand that it is possible to create a business strategy for the target adopters to capture the value by versioning based on token usage. The logic behind this is that not every organisation has the same number of patients and gives the organisations to pick the version that suits them.

Throughout the project I learned that you cannot just push out the product you created without validation. The created prototype should also be shown to ‘real’ customers to validate the willingeness to buy. If a prototype is not validated for willingness to buy, then there is no point in continuing the development of the product. In Careflow’s case, we validated that willingness to buy is high and real.

To answer the question in the title:

“You disrupt the market by investigating the need, apply the appropriate technology, create an appropriate AI and business strategy, test your prototype with your collaborators and validate the product with real customers”.

Thank you for sticking until the end and I hope you enjoyed the read!
Danny