Note: This post is excerpted from our report, The agentic AI workforce is coming to healthcare.

The AI industry is evolving rapidly, so it can be hard to keep up with all the changes and advancements to the technology, especially as they relate to AI agents. 

To help, we’ve compiled seven questions that can help you uncover critical details about agentic AI capabilities, support, security, and alignment with your business needs, reducing the risk of unforeseen issues should you look to bring AI agents onboard. Having answers to questions like these at the ready can also go a long way in gathering internal support across your organization.

What tasks can the voice AI agents take on?

It’s best to start with the obvious to determine whether there’s a fit and if the platform can actually do what you need it to do. Thus, asking what tasks the voice AI agents can handle helps you understand capabilities, limitations, and overall fit for your needs. 

This question also reveals whether the AI can handle real-time conversations, escalate issues when needed, and integrate with your existing systems. Infinitus AI agents, for example, call patients, payors, and providers, completing a variety of tasks – with the ability to create new use cases based on customer need. But not all AI agent vendors can make calls across the healthcare ecosystem, and may be limited in the types of tasks they can complete. By exploring task capabilities, you can assess how well the voice AI agent aligns with your operational goals and whether it can truly reduce manual effort while improving the patient experience. 

Will patients trust the AI agent?

Healthcare consumers are more skeptical than ever. Thus, transparency, empathy, and authenticity are essential in building trust, especially when marketing specialty medications. Brands that can materially improve patients’ healthcare journeys will stand out, and AI can play a role. 

Here are some examples of how to build trust when employing an AI agent, which are all elements of Infinitus’ approach to the use of voice AI:

  • Transparency: The AI agent immediately introduces itself as an AI agent by saying, “Hi I am a digital assistant from [Brand name]” – it doesn’t try to hide that it isn’t a human. 
  • Clear communication: The AI agents should be clear and empathetic, building a connection with their human counterpart. 
  • Patient-centered content: The AI agent should be created to specifically anticipate the questions patients will ask and have the necessary answers pre-approved and ready.

At Infinitus, we are trusted by 44% of the Fortune 50, including many top life sciences manufacturers, for providing agentic AI solutions, with more than 5 million calls completed. Our AI agents are proven in engaging with humans and following guidelines 100% of the time.

What kind of models does the solution use?

The specific models in use by a voice AI solution can affect its performance, customization, and scalability. Proprietary models can be tailored to specific needs, offering better accuracy and control and benefitting from an organization’s unique knowledge and data. Third party models can also bring their own strengths and advantages in performance, flexibility, and deployment speed. 

Knowing this helps assess the voice AI platform’s capabilities, potential for innovation, and how well it can adapt to unique business requirements.

Our own AI platform employs over 100 models, from highly performant shallow models to custom fine-tuned and in-house audio and text models to large-language models (LLMs) from OpenAI and Google. We can thus apply the right model to the right use case to deliver customer results.

Is the solution secure, compliant, and bias-free, and how will I know patient data is safe?

HIPAA and SOC II compliance are table stakes for any AI platform to serve healthcare customers. Beyond this, it’s also important to ask about the security and data protection measures in place, and be aligned with your internal security and compliance leaders on what matters most. 

Certainly, you’ll need to understand how protected health information (PHI) is handled, and whether it’s used to train AI models. At Infinitus, for example, we never use PHI to train our models, and have procedures in place to ensure training data is de-identified and PHI is completely removed. 

Regarding bias, be sure to ask about the review processes in place, whether they’re AI- or human led, and how inaccuracies and potential biases are corrected.

As an example, here are some of the way Infinitus delivers on a safety-first approach to voice AI: 

  • Data quality: We ensure the data used to train AI models is accurate, complete, and representative of the population across diverse demographic and therapeutic areas. 
  • Model validation: We assess the model’s performance on de-identified data and ensure its reliability and generalizability. 
  • Ethical considerations: We evaluate potential biases in the data and algorithms, considering patient privacy, and addressing the ethical implications of AI-driven decision making. 
  • Regulatory compliance: We ensure that models used in administrative and clinical use cases adhere to relevant regulatory guidelines.

How do I know AI agents won’t share incorrect information, collect incorrect information, or even make things up?

This all comes down to the safety protocols and guardrails the voice AI platform has in place. A well-designed voice AI platform should have strict data validation processes, access only to approved sources, and mechanisms to prevent hallucinations or misinformation. 

Infinitus, for example, can guarantee 100% adherence to standard operating procedures. Our AI architecture creates a discrete action space for the questions, answers, and statements that our AI agents are allowed to make. Understanding a voice AI platform’s safeguards will give you confidence that the AI agents can be trusted to support both patients and healthcare organizations with accuracy and integrity.

Pacing is important in ensuring that humans who converse with voice AI don’t get frustrated and hang up. How well does the AI agent engage in a conversation?

Pacing is crucial for an AI agent to effectively engage with a human in conversation. Just like in human-to-human communication, a well-paced conversation allows both parties to feel heard and understood. For an AI agent, this means accurately detecting when a human has finished speaking before generating a response, avoiding awkward interruptions or delayed replies. 

When a voice AI agent demonstrates an understanding of conversational rhythm, it fosters a sense of connection and rapport with the human user, leading to increased trust and willingness to interact further.

At Infinitus, our voice AI agents have been optimized over the course of nearly 100 million minutes of conversation to have the right pacing. We continuously test, refine, and integrate the best ‘end of turn’ detection models to ensure that humans have an engaging and satisfying experience with our AI agents.

How easily can the voice AI platform integrate with existing infrastructure?

IT teams don’t have time to be integrators of scores of disparate products. Seamless integration minimizes disruption and ensures that the new technology works harmoniously with current systems. This question helps gauge the complexity of deployment, whether there’s a need for additional resources, and potential compatibility issues, which can impact implementation timelines and costs. Ensuring smooth integration also maximizes the return on investment by leveraging existing technology and data without requiring extensive overhauls or custom solutions. 

At Infinitus, we make integration easy for our customers. We provide ready integrations with leading EHRs and CRMs, and a range of options for data intake via API, CSV, or native integrations. With the Infinitus API, customers only need to integrate once and can then add AI agents and AI copilots as their business evolves.

If you’d like to learn more about how to make sure you’re confident in how to evaluate agentic AI for healthcare, read our full agentic AI report, explore our AI agent offerings, or reach out to us today.