What is Amazon Connect Health?
Amazon Connect Health is an AI-powered healthcare service that helps healthcare organizations reduce administrative and documentation burden across the patient journey. Its agents are pre-built and fully managed. You don’t train models or manage infrastructure.
Amazon Connect Health offers two independent families of agents, built for different audiences. This guide is organized around that distinction:
-
Patient engagement agents handle voice-based patient interactions for health systems and provider organizations.
-
Point of care agents generate pre-visit summaries, clinical documentation, and suggested medical codes for clinician and coder review, and are built for healthcare technology builders to embed in their own applications.
Important
Amazon Connect Health is not a medical device in the US. Ambient documentation, the point of care agent currently available in the UK, is registered as a Class I medical device in the UK.
Information generated by AI may contain mistakes and should be reviewed. Healthcare providers retain full responsibility for the accuracy of clinical documentation and the coordination of patient care.
Topics
Which agents are right for you?
Use this table to identify which family of agents applies to you. Everything else in this guide is labeled as applying to patient engagement agents or point of care agents.
| Patient engagement agents | Point of care agents | |
|---|---|---|
|
Built for |
Health systems and provider organizations |
Healthcare technology builders, such as electronic health record (EHR) vendors and independent software vendors (ISVs) |
|
What they do |
Verify patient identity and manage appointments through voice conversations |
Generate pre-visit summaries, clinical documentation, and suggested medical codes for clinician and coder review |
|
How you access them |
Through your contact center |
Amazon Connect Health SDKs and APIs, integrated into your own application |
|
Contact center |
Required. Patient engagement agents are used through a Contact Center as a Service (CCaaS) solution. Amazon Connect Customer is natively supported. For other CCaaS solutions, contact your AWS account team. |
Not required |
|
EHR and data integration |
Epic EHR integration is natively supported through the Amazon Connect Health application for Epic, using Fast Healthcare Interoperability Resources (FHIR) R4 APIs and Epic private APIs. See Epic EHR integration. For other EHRs, contact your AWS account team. |
Clinical data is retrieved through FHIR APIs (AWS HealthLake or another FHIR server) and Amazon S3 |
|
Agents |
Patient verification, Appointment management |
Patient insights, Ambient documentation, Medical coding |
The following diagram shows how each family connects to your systems. Patient engagement agents are used through your contact center and work with your EHR. Point of care agents are embedded in your application and work with FHIR data and Amazon S3.
Patient engagement agents
Patient engagement agents handle inbound and outbound voice interactions with patients through your CCaaS solution. When a conversation needs a person, the agent escalates to your contact center staff with full context preserved.
| Agent | What it does | Status |
|---|---|---|
|
Patient verification |
Verifies patient identity at the start of a call by matching caller information against EHR records. |
Generally available |
|
Appointment management |
Helps patients schedule, reschedule, and cancel appointments through voice conversations, with optional real-time insurance eligibility verification. |
Preview |
Deep integration with Amazon Connect Customer
Patient engagement agents integrate deeply with Amazon Connect Customer: they are invoked from your contact flows, carry patient context across the call as session attributes, and escalate to your contact center staff through your existing queues with the escalation reason and full conversation context attached. Your staff see that context (demographics, verification status, appointment intent, and what the agent already completed) in a unified patient profile within the Amazon Connect Customer Agent Workspace. For details, see Sample contact flow and Patient profile.
Note
This integration is a patient engagement capability and is not part of point of care agents.
Point of care agents
Point of care agents assist clinicians and coders across the visit (before, during, and after) by providing relevant information and supporting clinical documentation and coding. You integrate them into your own application through the Amazon Connect Health SDKs and APIs. They do not require a contact center.
| Agent | What it does | Status |
|---|---|---|
|
Patient insights |
Generates pre-visit summaries that consolidate patient history, medications, and recent lab results, retrieved through FHIR APIs (AWS HealthLake or another FHIR server) and Amazon S3. |
Preview |
|
Ambient documentation |
Captures patient-clinician conversations in real time and generates structured clinical documentation for provider review. |
Generally available |
|
Medical coding |
Generates International Classification of Diseases, Tenth Revision, Clinical Modification (ICD-10-CM) diagnosis codes and Current Procedural Terminology (CPT) procedure and service codes, with modifiers, from clinical documentation. Each code is linked to the evidence that supports it. |
Gated preview |
Model training, evaluation, and guardrails
The agents described above are built on task-specific models rather than general-purpose language models prompted at inference time. This section summarizes the methodology that applies across the agents: how the models are trained, how their outputs are evaluated, and what guardrails are applied at inference. It is a summary; the full methods and results are reported in the reference cited at the end.
| Methodology | |
|---|---|
|
Training |
Models are adapted to clinical tasks by supervised fine-tuning on large-scale de-identified real-world clinical data, followed by reinforcement learning fine-tuning. Reward objectives are decomposed and optimized in sequential stages rather than combined into a single reward. This domain-adaptation approach is designed for high task-specific accuracy while controlling latency and inference cost. |
|
Evidence mapping |
Each generated proposition is aligned to the span of source material (conversation transcript, clinical documentation, or patient record) from which it was derived. Coverage of this alignment is itself an evaluated quantity: the fraction of generated statements supported by a source span is measured against real data. Evidence mapping provides per-statement provenance and is the mechanism on which downstream human verification depends. |
|
Evaluation |
Outputs are assessed along multiple dimensions using a combination of automated metrics, LLM-as-a-judge rubrics (automated scoring by a language model against defined rubrics), and blinded manual preference review by clinical specialists. Where a task admits a ground truth, precision, recall, and F1 are computed against it. Model releases are gated on adherence to clinical guidelines as assessed by specialist review. |
|
Guardrails |
Generated content is filtered for toxic or harmful output and validated against the expected schema, with non-conforming outputs rejected. Where an agent interacts directly with a person, generation is bounded by a constrained state architecture with explicit transitions, protections against prompt and system injection and against misuse, and defined escalation paths to human staff. |
Evaluation scope. Evaluation results are produced on internal, proprietary datasets and are not independently reproducible from public data. Performance in a given deployment depends on the characteristics of that deployment’s data and workflows.
For the full description of training methods, evaluation datasets, and quantitative results, see How We Built Healthcare AI You Can Trust: The Science Behind Amazon Connect Health
Accessing Amazon Connect Health
How you access Amazon Connect Health depends on the family of agents you use.
| Access method | |
|---|---|
|
Patient engagement agents |
Configure and manage agents in the Amazon Connect Health console, a web-based interface. You can also use the AWS CLI and the AWS API. |
|
Point of care agents |
Integrate through the AWS SDKs and the AWS API from your preferred programming language. See Developing with the AWS SDKs and the Amazon Connect Health API Reference. |
For AWS CLI setup, see the AWS CLI User Guide.
Supported regions
Amazon Connect Health is available in the following AWS Regions. Availability differs by agent.
| Region name | Region code | Patient engagement agents | Point of care agents |
|---|---|---|---|
|
US East (N. Virginia) |
us-east-1 |
Patient verification (generally available), Appointment management (preview) |
Patient insights (preview), Ambient documentation (generally available), Medical coding (gated preview) |
|
US West (Oregon) |
us-west-2 |
Patient verification (generally available), Appointment management (preview) |
Patient insights (preview), Ambient documentation (generally available), Medical coding (gated preview) |
|
Europe (London) |
eu-west-2 |
Not available |
Ambient documentation (preview) |
Note
Preview and gated preview agents are subject to change and are not intended for production use. Gated preview agents additionally require access to be granted by your AWS account team.
Related services
Amazon Connect Health works with the following AWS services. Each is used by one or both families of agents, as noted.
| Service | Used by | Purpose |
|---|---|---|
|
Patient engagement agents |
Cloud-based contact center service that powers the voice channel |
|
|
Point of care agents |
FHIR-compliant data store that patient insights can use to retrieve clinical records |
|
|
Patient engagement agents |
Serverless compute used for insurance verification integration |
|
|
Point of care agents |
Object storage for patient insights input documents, and for ambient documentation and medical coding output |
|
|
All agents |
Key management service used for encryption of data at rest |
Pricing
Amazon Connect Health pricing is based on usage. Each agent has its own pricing model. For current pricing, see Amazon Connect Health pricing
Next steps
Start by identifying which family of agents applies to you, then continue with the following:
-
Which agents are right for you? – Revisit the comparison if you’re still deciding between patient engagement agents, point of care agents, or both.
-
Setting up Amazon Connect Health – Understand the prerequisites and configure your environment.
-
Patient engagement agents or Point of care agents – Go directly to the family that applies to you.
