Empathetic outbound voice agent. Identity-verified. Compliance-aware.
Outbound voice agent for collections. Identity verification gate before any account discussion. Outbound responses validated against a compliance rubric before TTS reads them.
No black box. Each step is a typed-frame node you can edit, monitor, and replace.
The agent dials. The customer answers. The streaming-identity-verification-gate runs the verification script - DOB, account-last-four, security question. Pass = unlock account context. Fail = polite end.
Verified, the LLM has access to the customer's balance, recent transactions, and current arrangement options.
Every outbound reply runs through streaming-response-validator against your compliance rubric. Disallowed claims (e.g. legal threats, unsupported amounts) are caught before TTS.
On call end, the saver writes a full audit row: verification outcome, every rule that fired, every disposition offered.
Streaming graph contract, observability, and cost discipline come standard. The agent ships with a full test suite that runs in CI on every node version bump.
streaming-identity-verification-gate runs the verification script before account context is exposed. Configurable challenge order, retry budget, and lockout behaviour. Compliance-friendly by design.
Every LLM response goes through streaming-response-validator with your compliance rubric. Unsupported claims are blocked before TTS - never spoken to the customer.
The system prompt is tuned for respectful, non-aggressive collections language. Tone is enforced at every turn by the response-validator; the rubric ships with empathy guard-rails.
Conversation memory carries the verification result, the customer's stated arrangement preference, and prior turn dispositions. The agent doesn't re-ask for what it's already been told.
On call end, the saver emits one row per session containing: verification outcome, every rule fired, every disposition offered, every customer commitment. Regulator-ready.
Verification challenge order, response-validator verdicts, and disposition-emitter outputs land on the call's span tree. The canvas overlay shows what happened, why, and when.
Every dependency is a registered Strongly service or a model you control. Swap any one of them in the install wizard. The graph stays intact.
The marketplace template is the graph. Every customisation below is a config change or a single-node addition - never a rewrite.
Edit the gate's challenge config - add KBA, biometric voiceprint, OTP. The graph stays the same; only the gate's config changes.
Edit the response-validator's rubric. Common additions: state-specific disclosure language, FDCPA / regional-equivalent guard-rails, do-not-call-list checks.
The disposition-emitter ships with arrangement defaults (PIF, settlement, payment plan, hardship). Edit the catalogue per portfolio or per region.
Add streaming-handoff-detector. Verified disputes, hardship signals, or compliance-flagged turns escalate to a human agent without ending the call.
Audio recording is mandatory in some jurisdictions, prohibited in others. Flip streaming-recorder on or off per region in the wizard.
We don't leave until it runs. Talk to a forward-deployed engineer about deploying Collections Bot into your environment with your STT, your LLM, your TTS, your data.
Schedule a Demo