AI in healthcare is a classic case of putting the cart before the horse. Yes, this one:
Transformation can begin only if there is a felt need. If there’s none, you can roll off hype straight from the consultants marketing pitch (or sales deck). What is that isn’t working? What sort of system’s research is required to state the truth hiding in plain sight? Healthcare has become too bureaucratic and needs to be responsive to the felt concerns of its constituents.
AI in healthcare is a chicken-and-egg situation. Healthcare enterprises are not adopting it due to either lack of understanding or absence of “minimum viable product” to demonstrate efficiency of processes. For example, electronic medical records (EMR). Have they shown any demonstrable benefit to reduce medication errors, or made any quantum jump to charge up clinical decision support systems? (I am only using this as an example but it requires careful calibration of historical controls to determine efficacy). The data pipes remain as messy as before, because users’ habits are tougher (e.g. using text shortcuts) than the clinical problems encountered in the clinic.
Therefore, for a healthcare start-up, I have specific inputs (based on what I have read and understood):
- Founders overfocus on customer’s “needs”-though most customers don’t know what the product should be. There’s an entire field of behavioural economics on predicting the irrationality of customers. As such, we only have incremental builds instead of a “breakthrough”.
- Start-ups are caught up in company-building before product fit needs a rational alignment. Identification of “what problem you will solve”is the most difficult first step. I witnessed a startup focused on robots for “hospital sanitation”. What’s wrong with the existing solutions?
- For any start up – they need to define existing workflows, and then identify the promise to help the user “automate” it by identifying the pain-points. It requires understanding the felt needs, instead of a top down approach to adapting to a new technological solution. It also marks the riskiest assumptions that determine the success or failure of a new proposed change.
- Always test your hypothesis before you get a technical person who needs “poof-of-the-pudding” rather than your “vision”, which might merely be vaporware.
- Always focus on actions as a value proposition to make it simple; complexity only adds more layers of confusion.
If you observe the 2×2 table in the embedded tweet, you’d realise most companies fall in the bottom right of the quadrant. The most successful enterprises harness internal idea pipeline to address their workflows. A clinician who understands user-interface, design principles, returns on investment, need for cybersecurity and privacy (in terms of federated data) is more valuable than hiring consultants who point the obvious elephant in the room.
The process is an art and not exact science. The success for digital transformation depends on intellectual honesty, rigorous thinking, and knowing when something is not working out. It is impossible to determine how market place (or policy) will evolve. It requires a sandbox to test scenarios and understand that the process might fail altogether.
I have worked through some places and have a “bird-eyes view” on how to achieve a product fit, because the pain-points are common across geographies. Whether it is administration focused on “quarterly results” or grant committees that wish to determine prior proof of “academic success” (through publications), ideas are sidelined. Therefore, hiring from outside the country (or institution) provides fresh perspectives on existing problems – staying uncomfortable is the key growth signal for me.
RadOncNotes 