Skip to main content

Wild Elm Agent

Overview​

The Wild Elm Agent communicates with patients across channels such as SMS and voice and can carry context between those interactions.

The agent connects to the clinic's operational systems to answer questions and perform real actions. It follows the clinic's scheduling, identity, insurance, and operational rules when doing so.

Capabilities​

The agent can help patients:

  • Schedule visits
  • View upcoming visits
  • Reschedule visits
  • Cancel visits
  • Confirm attendance
  • Respond to clinic-required reschedules and cancellations
  • Join and respond to waiting-list offers
  • Receive and reopen intake forms and consent forms
  • Verify insurance eligibility and retrieve available benefit information
  • Ask questions about the clinic
  • Share supported media, including photos, videos, and audio messages

The agent can also send automated communications, including:

  • Visit confirmations
  • Visit reminders
  • Clinic-required reschedule notices
  • Cancellation notices
  • Intake form and consent reminders
  • Waiting-list offers

Available capabilities may depend on the clinic's configuration and integrations.

Identity and authorization​

Wild Elm verifies a patient's identity before revealing protected patient information or making changes to an existing patient record.

By default, Tier 1 verification uses:

  • First name
  • Last name
  • Date of birth
  • Additional identifying information when needed to disambiguate matching records

Verification applies to a specific patient record. It does not establish that everyone using the same phone number is the same patient.

This allows shared phone numbers, caregivers, and family members to use the agent without one patient's verification automatically granting access to another patient's information.

A failed verification also does not cause Wild Elm to assume that the person is a new patient.

Replies to automated messages​

Automated messages carry narrow authorization context.

For example, when Wild Elm sends a confirmation for a specific patient's visit, a reply to that message can remain associated with that patient and that visit. This allows a patient to respond naturally:

Clinic: Please confirm your visit tomorrow at 2 PM.
Patient: Can we do Friday instead?

Wild Elm can understand that the patient is asking to reschedule the visit referenced by the message.

This authorization is limited to the visit associated with the automated message. It does not grant access to unrelated visits or to other patients who may share the same phone number.

Normal identity verification applies when the conversation moves outside that scope.

Scheduling safety​

Wild Elm treats scheduling changes as transactions against the clinic's scheduling system.

The agent follows several safety guarantees:

  • It does not change a different patient's visit because they share a phone number.
  • It resolves which patient and which visit an operation concerns before changing an existing visit.
  • It does not treat an availability search or an offered time as a booking.
  • It does not report a visit as scheduled, rescheduled, cancelled, or confirmed until the corresponding operation has succeeded in the clinic's system.
  • It does not cancel a visit from an ambiguous request. Cancellation requires explicit patient intent and confirmation.
  • It respects scheduling constraints such as provider, location, visit type, insurance, room, equipment, and integration capabilities when those constraints are configured.

If Wild Elm cannot safely determine what should happen, it requests human attention instead of guessing.

Insurance eligibility​

Wild Elm can verify insurance eligibility through supported eligibility integrations.

The information available depends on the payer and the patient's plan. When returned by the payer, Wild Elm may retrieve information such as:

  • Coverage status
  • Copays
  • Deductibles
  • Coinsurance
  • Benefit details

Wild Elm distinguishes an authoritative eligibility result from cases where eligibility could not be determined. An unavailable, incomplete, or inconclusive eligibility response is not treated as evidence that the patient has no coverage.

Forms and consents​

Wild Elm can send patients forms and consent packets associated with their visits and remind them about outstanding forms.

Patients complete forms through the clinic's configured forms provider. Wild Elm can provide or refresh the appropriate form link and report completion status available from that provider.

The conversational agent does not independently claim that a form has been completed unless the forms system reports it as complete.

Human escalation​

Wild Elm requests human attention when it reaches a situation that it cannot safely or reliably resolve.

Examples include:

  • The patient explicitly asks for a person
  • Identity or authorization cannot be established
  • A required integration is unavailable
  • The outcome of a scheduling operation cannot be determined safely
  • The patient needs information that is not available to the agent

On asynchronous channels, requesting human attention does not necessarily end the conversation. The agent may continue helping when the situation becomes resolvable before staff intervention.

Context across channels​

Wild Elm can carry patient context between supported communication channels.

Cross-channel context is intentionally selective. A new interaction may receive a structured summary or relevant operational state rather than the complete transcript of another channel.

Context and authorization are separate concepts. Information appearing in an earlier conversation does not, by itself, authorize Wild Elm to expose protected patient information in a later interaction.

Wild Elm reuses patient-scoped information only when the current interaction has the authorization required for that information.

Operational boundaries​

Wild Elm operates within the capabilities and rules exposed by the clinic's systems.

If an integration cannot perform a particular action, Wild Elm does not represent that action as completed. When the system cannot safely continue automatically, the agent requests human attention rather than inventing an outcome.