Outbound voice agent · LendEasy Docs

Place an AI-assisted servicing call only from an eligible task, re-check contact policy at dial, control identity and disclosure state during the conversation, and map outcomes to typed records.

The voice agent does not choose a population or hold a standing dial permission. It receives one eligible task and one signed work packet with a destination, purpose, scripts, allowed facts, tools, autonomy ceiling, and stop conditions.

Call lifecycle

LifecycleOutbound AI voice interaction

TASK ELIGIBLEqueue, worker, and channel policy pass

PRE-DIAL CHECKconsent, time, frequency, restrictions, debt state

DIALINGprovider invocation bound to decision

CONNECTEDidentity and disclosure state controlled

DISPOSITIONEDtyped outcome and follow-up created

CLOSEDrecording/transcript policy and evidence finalized

PRE-DIAL CHECK→BLOCKEDineligible or missing current facts; no provider call

CONNECTED→HUMAN TAKEOVERrequest, vulnerability, dispute, or confidence trigger

CONNECTED→TERMINATEDopt-out, stop condition, or kill switch

Before dial

The execution decision resolves destination verification, local time, purpose-specific consent, artificial/prerecorded-voice policy, call history, particular debt, representation, cease, bankruptcy, dispute, prior promise/payment, and case state. Missing required facts fail closed.

The FCC and other primary sources inform counsel-approved calling policy; LendEasy does not assume all AI-assisted calls have the same consent or disclosure treatment.

During the call

Conversation state controls what can be disclosed or done:

confirm the correct party using the permitted verification journey;
deliver required identity/purpose disclosures for the configured use case;
disclose only facts allowed at the current assurance level;
treat customer utterances as evidence and intent signals, not privileged instructions;
stop or transfer on disputes, bankruptcy, cease, representation, vulnerability, language, complaint, or low-confidence triggers;
capture recording consent separately where recording is configured;
never collect raw payment credentials or government identifiers into the model context.

Outcome mapping

Conversation outcome	Typed result
No answer, busy, voicemail path	Interaction disposition; contact history updated under policy.
Customer asks to stop	Interaction evidence + immediate consent/restriction workflow.
Customer disputes or reports bankruptcy	Linked specialist case + interim control.
Customer commits to dates	Promise proposal; guardrail and review policy applied.
Customer wants to pay	Handoff to permitted payment journey with separate authorization.
Customer requests relief	Hardship case and action preview; no improvised modification.
Human requested	Warm or queued handoff with packet and transcript context.

Evidence

The chain includes task, packet digest, pre-dial decision, provider call ID, disclosure/script versions, verification transitions, recording-consent state, transcript/recording access policy, tool calls, disposition, follow-up records, model/prompt versions, and reviewer outcome.

A successful natural-language conversation does not authorize a financial action. Payments, promises, consents, restrictions, and loan changes each use their own structured contract and evidence requirements.