Excitement around artificial intelligence isn’t contained to early stage startups alone. Established companies across a wide variety of industries are implementing AI to enhance their product offerings. 

Innovaccer — a Series E healthtech firm founded in 2014 — is one such company. In April, the San Francisco-based business revealed its conversational AI assistant, Sara, meant to help healthcare professionals ask questions of their datasets. What they get in return are “human-like” responses. 

“What we have been doing is figuring out — what are the biggest time suckers for healthcare workers and how can we solve that with generative AI?” Kanav Hasija, Innovaccer co-founder and chief product officer, told 500 Global. 

Next week, Innovaccer (a 500 Global portfolio company) is scheduled to announce several more product launches at “The Innovation Keynote” virtual event. Hasija said that providing healthcare employees more easily accessible company data, while cutting down on documentation and research workloads, are the “two cornerstones” of Innovaccer’s upcoming releases.

But how has a nearly 10-year old company with over 1,000 employees adopted AI in the healthcare industry? 

Here are four keys that may have led to positive outcomes at Innovaccer thus far.  

Take the time to train your models.

Hasija said foundational models today know general knowledge, but often don’t know the intricacies of specific industries, like healthcare. For instance, ask a generative AI model for the definition of diabetes, and it’ll produce an acceptable answer. But, ask it to identify diabetic patients within a dataset, and it’ll run into problems. 

That’s why, over the past six months, the Innovaccer team has been training its models on healthcare data, formulas, and concepts to — as Hasija puts it — cut down on “hallucinations” and provide reliable insights. 

Many people are new to prompt writing. Remember that.

Popular AI tools like ChatGPT are designed with open text-box interfaces, which require users to write their own prompts. The problem with such a design, Hasija said, is that to get to the “best answers,” a certain level of expertise is needed. 

“We can’t say — Doctors, you’ve learned medicine. Now learn prompt engineering,” he said. 

Instead, Innovaccer’s new product designs replace the open text box with a series of questions, which help to generate prompts for the user automatically on the backend. 

User feedback is essential.

Companies can create AI models, but the ones who will learn “better and faster,” are the companies that build more robust feedback loops into their products, Hasija said. As he put it — a simple thumbs up or thumbs down for users to indicate whether an answer is helpful isn’t enough. 

One idea, Hasija said, is to build a feedback loop that understands exactly what part of a response the user edits before implementing it into their workflow. 

Clarify your internal approach to AI.

At a large company, the question may arise whether every team, or only certain teams, should be working on AI-centric products. 

Hasija said Innovaccer took somewhat of a hybrid approach. For training its foundational healthcare models and creating its internal API wrappers, the company designated centralized teams. 

From there, it’s given product teams across the organization freedom to leverage its AI capabilities for their specific focus areas where they see fit. 

Interested in learning more about Innovaccer’s new AI products. Check out their Innovation Keynote from Tuesday, August 22nd.