We are a venture fund focused on investing in companies developing technology that can make healthcare more affordable and accessible. Increasingly, AI forms the foundation of this technology. We seek dual returns from our investments: both societal and financial. If customers and societal stakeholders, such as patients and insurers, don’t see value in the technology, it becomes difficult for the company to achieve profitability.

One of our partners, Noaber, invested in a company that uses patient data from hospital information systems (HIS) to estimate the risk of sepsis in patients. Studies show that around 6% of ICU patients experience sepsis, with mortality rates ranging from 28% to 55%. Early detection of sepsis allows for timely antibiotic treatment, significantly reducing mortality.

We have also invested in a company developing software that will improve the ability to predict whether immunotherapy will be effective for cancer patients. Immunotherapy can have serious side effects, and many immunotherapies benefit only a small subset of patients. This AI application, through techniques like massively parallel sequencing and subsequent data interpretation, can analyze whether a patient will respond to immunotherapy. If not, the treatment—and its side effects—can be stopped early. This more personalized approach to administering expensive medications can also help control healthcare costs.

When selecting companies to invest in, we look for solutions that meet practical needs. A company must be viable, meaning there must be a market for its technological solution. The technology must also contribute positively to healthcare, ideally at the same or reduced cost. However, we can’t always prevent a technological solution from increasing costs. We don’t control how customers use the technology we’ve invested in. If they treat the solution as an add-on rather than a core tool, there’s little we can do to change that.

We see many AI innovations designed to diagnose diseases or conditions. These are interesting for research purposes, but their practical implications aren’t always significant. For instance, a technology that detects early-stage dementia might be scientifically impressive, but with limited treatments for dementia, early detection has limited practical value. Additionally, few people visit their doctors suspecting early dementia, so the AI system lacks the necessary data to be truly effective.

There are many developments in AI, but often they are relevant to only a small group, limiting their broader impact. Scaling an innovative product is challenging and expensive. Scientific validation is required, and the entrepreneur must meet various standards to bring the product to market. From the outset, an entrepreneur must ask: What problem are we solving, and how large is this problem?

AI is particularly useful when analyzing large amounts of relatively homogeneous data, such as patient records for a specific condition. An algorithm can analyze thousands of patients based on a select number of characteristics—such as gender or inclusion criteria for a study—far better than a human can.

However, when it comes to diagnosing a single patient, such as one with a skin condition, AI can’t yet analyze every possible diagnosis and definitively conclude whether the patient has basal cell carcinoma, melanoma or another condition. AI might suggest the patient has melanoma, but that might not be the patient’s primary concern. The patient wants to know what’s wrong with their skin and what treatment options exist. When it comes to answering those specific questions, humans are still better than AI.