This episode reveals how AI voice agents are not just automating tasks but systematically exposing and correcting deep-seated inefficiencies within healthcare's legacy communication infrastructure.
An Introduction to Infinitus and its Mission
- Infinitus, co-founded by Ankit Jain, focuses on using AI voice agents to solve the critical workforce shortage in healthcare. The platform aims at automating time-intensive voice interactions, augmenting both administrative and clinical teams.
- Large Language Models (LLMs): These are advanced AI models trained on vast amounts of text data, enabling them to understand, generate, and process human language. Ankit notes that LLMs are integral to every part of the Infinitus stack, from speech-to-text and natural language processing to speech generation.
- Ankit emphasizes that Infinitus was founded in 2019, well before the current hype cycle around voice AI, positioning them as early pioneers in the space.
The "Bruce Willis" Test: Proving AI's Viability
- Ankit recounts the company's very first, pre-incorporation test to prove the viability of their idea. The team built a demo agent to call United Healthcare and perform a simple benefits verification for a fictional patient, "Bruce Willis."
- The first attempt failed when the agent on the other end hung up. The second attempt, however, was a breakthrough.
- Although the agent couldn't find the fictional patient, the human on the other end engaged in a standard verification dialogue, asking for a date of birth.
- Ankit Jain describes this as the critical moment: "That was proof point number one that it is possible to automate a call between a machine and a human... and that was the jumping off point to say let's build a company around this."
- Strategic Insight: This initial, low-stakes experiment demonstrates a core principle for AI innovators: validating human-AI interaction in a real-world environment is the most crucial first step, even before building a full-fledged product.
Scaling with Guardrails: LLMs vs. Small Language Models
- Infinitus has scaled significantly, now handling over 5 million phone calls and 100 million hours of audio. Ankit clarifies that this scale is achieved not just with LLMs, but with a hybrid approach.
- Small Language Models (SLMs): These are more compact, specialized AI models trained for specific tasks. Infinitus uses SLMs to create "safe, compliant, guardrail conversations."
- This dual-model architecture is a key risk mitigation strategy against AI hallucinations (when an AI generates false or nonsensical information), which is critical in a high-stakes field like healthcare.
- Ankit notes the challenge of user perception, where technology is expected to be perfect, even when its human counterparts are not. The goal is to be "significantly better than human counterparts."
From Back Office to Proactive Care
- The company began by targeting administrative back-office functions like benefit verifications and prior authorization status checks, where process friction causes significant patient drop-off and avoidable medical costs.
- The long-term vision is to create a proactive healthcare system, moving away from the current reactive model.
- Ankit draws an analogy to a "pizza tracker for healthcare," where patients are kept informed at every step of their care journey, reducing anxiety and improving transparency.
Meeting the System Where It Is: The Multimodal Approach
- Ankit stresses that transforming healthcare requires meeting all stakeholders—patients and health systems—on their preferred communication channels. A one-size-fits-all approach is doomed to fail.
- The platform must be multimodal, supporting voice, text, and chat to accommodate varying patient preferences and needs.
- He highlights the emotional state of a patient after a scary diagnosis, noting they often think of critical questions late at night. An accessible, multimodal AI can provide answers and support in those crucial moments.
A Strategic Reversal: From Autonomous Agents to Co-Pilots
- In a move that reverses the typical product roadmap, Infinitus started with fully autonomous agents and later introduced an AI co-pilot product.
- Autonomous Agents: These are AI systems that complete tasks end-to-end without human intervention (e.g., making a full verification call).
- AI Co-pilots: These tools assist humans in real-time. The Infinitus "FastTrack" co-pilot handles navigating automated phone menus (IVRs) and waiting on hold, only bringing in the human employee when a person on the other end is ready to talk.
- Strategic Implication: This dual offering shows a deep understanding of the market. Some tasks are ripe for full automation, while others require a human touch, but with AI handling the low-value, time-consuming portions.
The Inevitable Shift: The Future of APIs vs. Phone Calls
- Ankit expresses a hope that most back-office communication will eventually transition to APIs (Application Programming Interfaces)—protocols that allow different software systems to communicate directly.
- He envisions a future where phone calls are reserved for richer, more complex interactions, while routine data exchange happens digitally.
- Patient-facing communication, however, will likely remain multimodal, dictated by patient preference.
The Accidental Feedback Loop: Building Payer Partnerships
- In a revealing anecdote, Ankit shares that for the first three years, his personal cell phone was the designated callback number for the AI agents. This provided an unfiltered feedback channel.
- This led to supervisors from major payer call centers calling him directly, asking, "Hey, who are you and why is your bot calling us?"
- These conversations became opportunities to explain the AI's legitimacy and efficiency, leading to formal partnerships. Some payers have even set up dedicated call centers staffed with agents trained to interact specifically with the Infinitus AI, "Eva."
- This demonstrates a powerful, emergent go-to-market strategy where the product's usage itself creates partnership opportunities and a network effect.
When Machines Talk: The Reality of Bot-to-Bot Communication
- The host asks the "killer question" about whether the Infinitus bot ever talks to another bot. Ankit confirms it happens.
- He explains this counter-intuitive reality: for some large enterprises, updating their IVR (Interactive Voice Response) systems is organizationally easier and faster than building and securing a new external API.
- These enterprises will enable their IVR to provide more data specifically when it recognizes a call from an Infinitus number, creating a machine-to-machine conversation over a voice channel.
Navigating the LLM Revolution: A Model-Agnostic Strategy
- Ankit discusses the challenge of building on a rapidly evolving foundation model landscape.
- From day one, Infinitus built a "discrete action space"—a system of guardrails that maps the LLM's understanding onto a pre-approved set of safe actions. This was crucial when early models (like T5) were prone to hallucination.
- The company maintains a model-agnostic infrastructure, allowing them to "rip and replace" and even use a concert of different models to achieve the best results.
- Actionable Insight for Researchers: This highlights the importance of building a flexible abstraction layer over foundational models. The true, defensible value lies not in the underlying model, but in the proprietary data, fine-tuning, and safety guardrails built around it.
Building the Team: Passion Over Pedigree
- When hiring, Ankit's team screens for a passion for solving healthcare problems above all else. He finds that many talented individuals from other tech sectors (like ads and gaming) are actively seeking more meaningful work.
- He notes that the democratization of AI tools is making it easier to build valuable applications, shifting the focus from raw model-building talent to those who can apply the technology to solve real-world problems with proprietary data and deep workflow integration.
The Go-to-Market Moat: Why Big Tech Can't Just Copy-Paste
- Ankit addresses the question of why a large tech company couldn't easily replicate their business. His answer centers on the messy reality of healthcare operations.
- Every health system, and even every team within it, has slightly different workflows and standard operating procedures.
- The defensible moat is built by "meeting the customers where they are," which requires deep, custom integrations into existing workflows and using proprietary data to tune models for those specific environments. The value is in the last-mile delivery, not just the core technology.
The Evolving Buyer: From Skeptical Ops Leaders to AI Review Boards
- The sales process has evolved dramatically over five years:
- Early Days: Selling "black magic" to skeptical operations leaders, requiring extensive pilots and bake-offs to prove the technology was real.
- Post-ChatGPT Hype (2022-2023): A shift to customers asking, "What else can you do?" leading to a race to expand features.
- Current State (2024): Large enterprises are now putting on the "brakes," establishing AI review boards to vet the influx of AI vendors for bias, data security, and compliance.
The Bake-Off: Uncovering Systemic Inefficiency
- During early pilots, Infinitus compared its AI's results against the client's human team. They found a 25% discrepancy rate. Initially, this was perceived as AI error.
- However, when they had the AI call the same payer twice for the same patient, they still found a 25% discrepancy, revealing a fundamental inconsistency in the human-powered system.
- This insight allowed Infinitus to build a knowledge graph to identify and challenge incorrect information in real-time, turning the AI into a quality assurance tool.
- Investor Insight: This is a prime example of how AI can create value not just through automation, but by surfacing and correcting costly, hidden inefficiencies in legacy systems.
A New Operational Muscle: Beyond the Tech Budget
- Ankit confirms that Infinitus is consistently categorized under a customer's operational budget, not their technology budget. The platform is viewed as a way to augment staff and scale operations elastically.
- A key example is the "blizzard" of calls in January, where providers verify patient insurance for the new year. Instead of hiring temporary staff for months to handle a few weeks of work, they can use Infinitus.
- This positions the product as a solution to labor shortages and a driver of revenue growth, rather than a simple IT cost.
The Emerging AI Ecosystem: Convergence or Divergence?
- Looking at the broader healthcare AI market, Ankit sees an open question about how different "killer apps" (like AI scribes, RCM tools, and voice agents) will evolve.
- He notes a coming conflict: "The agent companies are going to try to go down, the platform companies are going to try to go up."
- It remains to be seen whether a single "super app" or orchestration layer will emerge, or if the market will remain a collection of specialized, best-in-class solutions.
The 5-Year Horizon: The Power of Context-Aware AI
- Ankit believes the next major leap will come from aggregating currently disparate and missing data, such as Social Determinants of Health (SDoH)—non-medical factors like housing and food security that impact health outcomes.
- As AI makes data collection and aggregation easier, LLMs will act as powerful "translation engines" to standardize this information.
- This will unlock a future of truly personalized, context-aware communication and care that will make today's system feel archaic.
Conclusion
This discussion shows AI's true value is in navigating and improving complex human systems, not just replacing them. For investors and researchers, the key is to identify companies building defensible moats through deep workflow integration and proprietary data, turning systemic friction into a competitive advantage.
