Salesforce announced the launch of Agentforce for Health, an AI-driven platform that automates tasks like appointment booking, benefits verification and clinical trial matching. The tech giant is also partnering with healthcare technology company Athenahealth to streamline workflows, such as summarizing patient histories and processing prescription refills.

Amit Khanna, senior vice president and general manager of Salesforce, sat down with MobiHealthNews to discuss the making of Agentforce for Health and highlight what its partnership with Athenahealth will offer clinical teams.  

MobiHealthNews: Can you tell me about Agentforce for Health?

Amit Khanna: We are bringing a library of agents, as we call it Agentforce for Health agents, and it is a set of topics and actions that improve efficiencies in healthcare. So, whether you look at access to care, appointment booking, benefits or verifications, things that take time but ultimately do not give time for the provider or the care teams to be in front of the patient, but need to be done.  

So, making sure we simplify and make those processes efficient through our Agentforce platform. So, bringing Agentforce for Health is the one big piece that we are doing, which includes this, plus life sciences auto-matching of patients to clinical trials and also in public health space with home health, where we see, if you look at areas outside U.S., also in home health, they get funding from the government, and they need to find services, and this is a very manual process today. So, we are automating that with Agentforce.

MHN: What are the partnerships Salesforce is announcing?

Khanna: We are announcing three main partnerships. One is with Athenahealth, which is a big EMR in the outpatient space. So, we will bring Athenahealth and Salesforce's Agentforce together in simple scenarios like patient visit summaries. 

If you look at Athenahealth, there could be a big discharge summary, and now with Agentforce, with one click, when the member calls, the patient calls to the call center, we can have a summary right in front of the agent, without their doing swivel chair trying to find out what happened. It gives similarly missed appointments, care gaps, and everything we can connect to Athenahealth, and brings it up front in the call center when you call your provider. 

So, it might be your calling for, "Hey, can I do a prescription refill?" Today, you have to swivel a chair, but now, with Salesforce Agentforce, we can just on the Agentforce window, say, order a refill, and with our partnership with Athena, it will land in Athena as a pharmacy order. If Athena queue is a yes, it can be auto-approved, and it will land at their nearest pharmacy, which is preferred by the patient. So that's the level of integration we are doing with Athena.

MHN: Salesforce works a lot in healthcare. How does Agentforce enhance its work in the sector?

Khanna: I would like to first give a difference of generic LLMs. So, as you know, the whole world is about hype with LLM, and AIs will solve the problems. So, yes, we do agree to that, but also AI should be in the flow of work, which is where we come into play. If you look at most of our health plans in the U.S., most of our providers in the U.S. run Salesforce, for example, in a call center. 

Now, how can agents or AI help? That's where we are bringing these two things together – flow of work, understanding of the business process, which we know of as the workflows, and making sure your generative AI is trusted, is bound by the data that you know of and is not giving answers from the internet, basically. 

So, today, if you are, let's say, a health plan that serves members at risk, normally, what you will do is, once they're discharged from the hospital, you will call them and do an assessment. "How is your health? How was your experience?" Med management, reconciling meds. Before you do that work, normally, these care managers have to get an insight into what happened. Let's say, if it's me, what is my medical history? I was in the hospital. Am I living alone, or am I with family? So, they try to gather this information, which today takes approximately six minutes to prepare for that call. 

Now, imagine a world in which an agent sitting right in that call center application and a human can click and say, summarize Amit's past medical history. Thirty seconds, it goes into Salesforce, it goes into Athenahealth, gets that information, summarizes it, and without doing any clicks, swivel chair, you have a nice summary that you can read. At best, this work will take you one-and-a-half minutes, compared to six to seven minutes. So, for every call, you have saved five minutes. That's the place we are going towards.

MHN: How do you ensure that AI hallucinations are not present within these summarizations?

Khanna: I'll give you exactly how we make it happen. So, in the patient summarization, we are not asking the agent or the AI to randomly pick up content and then summarize it. We are actually guarding it and telling what we call topics and actions, saying when you have this topic of patient summary, you will only do these three actions. You don't have to think. So, we are removing from the AI, the thinking part of what actions I need to take, and we are also grounding it with the data that it should have access to. 

So we are applying the security and sharing rights that Salesforce is known for, and then we allow the agent to go into those datasets with that trust, bring that information back, and when we send it to an LLM model, we actually remove your PHI [protected health information] to create that summary. 

Lastly, which is very important, we are not doing it without a human in the loop because it's new technology. There will be places where it might hallucinate, even after all the effort we have put in, so we get the summary, but we also have the human so that a human can verify that yes, it is correct, and they don't give the wrong advice and they can have a feedback loop, which says, "Yes, this is okay" or "This is not okay." So, yes, we are also learning, but we are putting a lot of safeguards in place to make sure we don't create challenges for our patients and members.