On the importance of being guided by the customer needs over exciting technological challenges
In the workday of a scientist and programmer it is easy to get drawn into the excitement of improving model metrics or solving bugs, but constant focus on detail can distract one from the important bigger picture. It is in this mindset that we tend to ask questions of “can we do x”, when it would be more appropriate to ask “should we do x”.
Autoscriber recently took part in the LifeSciences@Work NWO Venture Challenge, a series of bootcamps and coaching sessions designed to mature a breakthrough idea into a solid business case within the life sciences domain. Facilitated by Math Kohnen and Chrétien Herben, both seasoned entrepreneurs and life sciences experts, we left our comfort zone of clinical Natural Language Processing (NLP) to be challenged on customer value propositions, validation de-risking strategies, key assumptions, financial models, and business plans.
When challenging our understanding of typical competitors during the first bootcamp, we were presented with the following analogy:
When you are in the business of making mouse traps, your competitors are not just other manufacturers of better or faster mouse traps, but cats as well. The customer is buying a solution to a mouse problem, not a mouse trap.
This analogy is also applicable to our case of getting lost in details. If one fails to remember that you are selling a solution to a mouse problem, it becomes tempting to over engineer springs and levers. And while this could be a satisfying area of research and development for a mechanical engineer, it makes little business sense in this context of mouse traps:
Mouse traps for physicians
One of Autoscriber’s main goals is to reduce the administrative burden of physicians. We are developing a real-time AI solution that listens in on a doctor-patient consultation and provides accurate and structured data capture of key information such as medicine, dose, side effects, symptoms, conditions, and treatments. These structured elements, once approved by the physician-in-the-loop, are then synced with the Electronic Health Record. This approach reduces the time spent writing medical summaries, and the structured data are important for future time-savings as they provide an immediate overview of a patient’s record.
In researching and developing the various components of this complex system, we frequently have had to step back and assess whether we are focussing effort in the right direction for maximum returns. While a 1,5% improvement in F1 score for Named Entity Recognition is a positive outcome from a data scientist’s perspective, how much of this improvement is felt by the physician? Would it be more valuable to develop intelligent systems to help automate the referral process or requests for diagnostic tests?
Armed with these lessons on keeping the focus on the customer and the ultimate value proposition, we’ve adopted a few key practices at Autoscriber:
- Close contact with the customer — We have regular feedback sessions with various stakeholders in partner hospitals to soundboard ideas, and we plan sessions where we can simply watch doctors interact with our software to learn more about their needs
- Quick iterations — Push to production frequently, so we don’t waste time perfecting features that aren’t useful
- “Zoom out sessions” — force even the biggest tech nerds to learn about what’s happening in the industry and who our customers are
- Radical honesty — we aim to let everyone in the team feel empowered and encouraged to say “hey, this doesn’t solve the core issue, let’s kill it”
For more articles on applying AI in the healthcare sector, follow us at Autoscriber:
P.S. We’re hiring!
