Scaling patient outreach inside a health system has never been a matter of simply “having a conversation.” Real progress depends on making sure the right actions happen at the right moments and doing so consistently across thousands of patients with varied needs. Tom’s AI agents were designed around this reality. They call and text patients, pull in human support when needed, capture the details that matter, and send structured information back into clinical workflows where teams can use it.
This spotlight explains how Tom works: how agents coordinate tasks across voice and text to meet patients where they are, the skills that support real clinical workflows, and the role long‑term memory and disciplined data handling play in creating a predictable experience that improves engagement and enables more continuous care.
Beyond Talking: How Tom Drives Follow‑Through
Tom uses modern AI models, but what differentiates it in healthcare is not the model; it’s the system built around it.
Tom is powered by a purpose-built orchestration layer designed to reliably take action in complex, real-world care workflows.
When a patient raises an issue that requires staff involvement, Tom doesn’t just suggest next steps – it intelligently routes and transfers the call using context-aware logic. When a patient needs to reschedule, Tom doesn’t send a generic link – it generates the right pathway based on the patient, provider, and workflow. When education is needed, it delivers approved, governed content across channels. And when patient-reported information is shared, Tom captures it in structured, usable data that feeds downstream systems.
The platform is built to act, not just to talk. It’s more than a response generator; it’s a command center that coordinates actions across the complex web of systems and workflows that healthcare runs on.
That commitment to follow-through is what truly matters. In an environment where reliability isn’t just a feature but a requirement, it’s the difference between a promising idea and a solution that works in the real world.
Multi‑Modal Agentic AI for Patient Outreach
Patients communicate across channels, and care teams need a system that can meet them wherever they are. Tom supports voice, text, and voicemail in a unified workflow. A patient may begin on a phone call, receive a personalized scheduling link over text while the call is still in progress, or get an educational video sent moments after the conversation ends. This flexibility ensures that each element of the workflow occurs in the format where it is easiest for the patient to act.
Texting enables slow‑burn interactions that naturally unfold over hours or days. Tom keeps context across these extended threads. If a thread goes quiet, the system can send a brief reminder, so the patient does not have to restart the interaction from scratch.
To reduce friction created by spam filters and unknown numbers, Tom can send a branded contact card during onboarding. Once saved, future calls show up as a known, trusted contact from the health system. This improves answer rates and reassures patients that the outreach is legitimate.
Multi‑modal support is not an add‑on. It is central to how Tom allows health systems to operate at scale, maintaining continuity and reducing the fragmentation that often occurs across channels.
Voice You Can Trust in Real Conditions
Outbound voice is one of the most challenging channels in healthcare because real conditions rarely resemble a controlled demo. Patients answer while watching television, standing in a grocery store, or driving with navigation systems running. Telephone networks compress audio, introduce delay, and vary widely by carrier. Voicemail systems behave inconsistently, and patients interrupt frequently.
Tom’s voice pipeline is built for this environment. The system maintains low latency, so it does not sound delayed or unresponsive. It handles interruptions smoothly so patients can speak naturally without the interaction breaking. It recognizes background noise and variations in audio quality typical of phone networks. Tom also detects voicemail accurately, leaving messages only when appropriate.
Tom supports extended hours, so health systems can reach patients when they are actually reachable. Evening and weekend availability increases completion rates for reminders, rescheduling, and chronic care check‑ins.
The result is a voice experience that feels stable in real conditions. The engineering is complex, but the patient experience is simple: a call that sounds clear, responds promptly, and adapts as real life happens.
Building Continuity into Patient Communication
Tom’s long‑term memory enables more consistent interactions across repeat outreach. The system stores only the information that meaningfully reduces friction and improves communication. Examples include how a patient measures blood glucose, whether they prefer calls or texts, and other preferences that shape how the system should engage.
This memory also extends to non-clinical characteristics that build rapport and drive engagement, much in the way that humans remember personal details about the people they care for. For instance, Tom can remember that a patient has a dog. This detail can be woven into a conversation to encourage positive health behaviors, such as taking more frequent walks to increase activity levels. It’s a small detail that makes the interaction feel more personal and connected.
When the patient interacts again, Tom uses this context to go straight to the relevant portion of the workflow. A patient who uses a continuous glucose monitor does not need to repeat that information before providing updated readings. This reduces unnecessary back‑and‑forth and makes each interaction feel more efficient.
Memory also prevents negative experiences. If a patient has said they do not wish to discuss a certain topic, Tom retains that preference and avoids raising it again. The system is not interpreting long‑term symptoms or diagnosing patterns. The goal is consistent, respectful communication grounded in what the patient has already shared.
This capability for persistent memory is the basis for more sophisticated applications. By remembering patient context, the system can go beyond simple interactions and manage more complex, multi-step processes.
The Pre‑Visit Workflow: A Coordinated Set of Skills
The pre‑visit workflow shows how Tom’s capabilities combine to support a complete patient interaction. It begins by confirming the upcoming appointment and gathering the patient’s agenda. This includes what they plan to discuss, any concerns they have been experiencing, and questions they want to raise with their clinician. Instead of following a rigid script, Tom adjusts based on the information the patient provides within the limits of the workflow.
If the patient cannot attend the appointment, Tom can send a personalized scheduling link that directs them into the correct self‑service flow. If self‑service is not possible or the patient prefers a human, the system routes the call to a scheduler. During the call, Tom can text educational materials that help the patient prepare for the visit, such as what to expect or how to get ready.
To ensure the information is where the care team needs it, Tom generates a structured summary and delivers it directly into the EHR after the conversation. This includes what the patient reported, any concerns requiring same-day attention, and a clear list of staff tasks. By extracting and organizing patient-reported data within the EHR, the system saves care teams from having to search through full audio transcripts.
The pre‑visit workflow combines conversation management, cross‑channel communication, rescheduling logic, education delivery, and post‑processing into a single, scalable service.
Where Decisions Come from and How They Are Audited
Tom does not decide which service should run next. That selection is made by the orchestration layer, Best Next Action. Once a service is chosen, Tom executes it using stored memory, the correct channel, and the skills defined for that workflow.
After each interaction, Tom’s post‑processing converts the conversation into structured outputs. Summaries, tasks, and patient‑reported information are extracted so care teams can act without reviewing raw audio. Auditing tools evaluate whether Tom followed the correct workflow and adhered to required rules. These insights feed a continuous improvement process led by product managers and prompt engineers.
What This Means for Patients and Care Teams
For patients, Tom provides a consistent, predictable way to interact with their health system. It remembers preferences, reduces repetition, supports both voice and text, and makes common tasks simpler. Scheduling, sharing readings, receiving education, or asking for a human can be done without navigating complex systems.
For care teams, Tom reduces the invisible work that accumulates around patient communication. Fewer voicemails require manual follow‑up. Fewer calls are misrouted. Summaries arrive structured and consistent. Patient‑reported data enters workflows in usable form rather than being trapped in transcripts.
The result is more reliable communication and fewer gaps where care can fall through. Tom handles the routine work and the data shaping, allowing teams to focus on the clinical care only they can provide.
