People who know me probably won’t be surprised when I say that I like to look for trends in what AI startups are doing.
While I see enormous commercial potential in AI, I also know it’s very tempting for people to make all kinds of too-good-to-be-true claims about big real-world problems that they plan on solving with it.
One of the most promising trends I’ve noticed is applying AI to very niche but nevertheless important problems. At this stage in the general evolution of AI, I think it currently works best to make well-defined, specific solutions.
Let me give you a great example of clever application of AI to a commercially valuable niche. It’s a product offered to healthcare facilities by Semantic Health, a firm that RiSC Capital has had the chance to work with in recent years.
A severe administrative headache that healthcare providers like hospitals face is billing the government and/or insurance companies. Every time a healthcare facility sees a patient, all the individual treatments it provides have to be assigned a billing code so they can be processed. Once you take into account that there are tens of thousands of these codes and that a hospital will typically treat tens or even hundreds of thousands of patients a year, you start to see why things get complicated.
The frontline care workers don’t do this assigning work, because their priority is obviously to treat the patient. Instead, using the records the frontline health professionals produce, a hospital will get a team of people known as “coders” to assign the codes to the treatments. This they do by combing through, line by line, the mountains of forms that the doctors and nurses fill in while they’re treating patients.
The process is very time consuming, and as you can imagine, there’s a lot of scope for human error and under- or overestimation. From a hospital’s perspective, the financial stakes of the exercise are enormous, and so it needs to get it right.
The Semantic Health team looked at this process and saw an enormous opportunity for AI natural language processing to help out. In particular, they grasped that because hospitals’ treatment records are repetitive in terms of their content, even if they tend to be very unstructured, as datasets they were suitable for the training and learning processes that good AI solutions depend on.
Semantic were sure an AI solution could make the coding process not only much faster but also more effective, in the sense that it could pick up on more billable items than human coders can. They set about proving all this in a way that anyone who’s trying to commercialize a deeptech idea should take note of.
The team approached a hospital and got permission to process a small sample of its records that had already been coded using traditional methods. That is, they put their idea head to head against the coders.
The AI found about half a million dollars of additional billable treatments. Given that the dataset amounted to just a fraction of the hospital’s annual output of records, it soon became very clear that Semantic’s solution could be enormously valuable to healthcare facilities. Unsurprisingly, hospitals that have heard about the product have been pretty enthusiastic about integrating it into their systems.
My hunch is that there are countless other niche real-world problems like hospital billing that are just waiting to be solved by AI startups similar to Semantic Health.
Investors who back those startups are going to be very happy indeed.