Collection Call Compliance Monitor

Collection Call Compliance Monitor

Audit 100% of your collections calls — AI and human — against the RBI Fair-Practices Code, and dial with a pre-certified compliant agent so a single breach never blacklists your agency.

Built on the Telenow App Platform as a pure declarative app: the data model, the scoring rules, the compliant agent, the call-log webhook, the escalation workflow, and the audit UI all run on-platform. No external server.

The problem (and the number)

RBI's tightened Recovery Agents guidelines and the 2025 Centralized Feedback Loop let a borrower's complaint propagate across lenders — a pattern of Fair-Practices breaches can get a collections agency de-empanelled / blacklisted, cutting off its book overnight. Yet compliance is measured the old way: a QA team spot-checks a tiny 1–2% sample of calls, and audited human-collector calls clear only ~87–92% of the rules. The other 98% of calls — where the out-of-window call, the "we'll send the police" threat, or the "I'll tell your office" third-party disclosure actually happens — are never scored. One of those, complained about, is enough.

Collection Call Compliance Monitor scores every call instead of a sample, on both your AI agent's calls and your human collectors' calls, and ships an agent that is compliant by construction.

How it works

AI agent call ─(call.analyzed event rule)─►  call_record (scored: false)
Human collector call ─(dialer/CRM → call_log inbound webhook)─►  call_record (scored: false)
        │
        ▼
   Audit queue  ──"Run audit"──►  getTelenow().ai.llm  scores each transcript vs the 5 RBI rules
        │                                   │
        │                       compliance_check (per rule)  +  violation_alert (per fail)
        ▼                                   │
   Dashboard: 100% coverage,        object.violation_alert.created
   compliance rate, per-rule                │
   breakdown, human-vs-AI          escalate_violation workflow
                                     ├─ delay ~2 min
                                     └─ outbound-call {{settings.compliant_agent_id}}
                                          → rings {{settings.escalation_phone}} (compliance desk)

The scoring is the App-AI path: getTelenow().ai.llm reads each unscored transcript against the five rules (billed to the org wallet via the ai:llm scope) and writes the verdicts. It is guarded — when the AI gateway isn't reachable (local demo, or no billable session), it falls back to a deterministic rubric scorer so the queue always clears.

The five RBI rules scored on every call

The rbi-rules knowledge base ships each rule as a scorable rubric (pass/fail criteria + example breach snippets) that both the LLM scorer and the bundled agent answer from.

Agent tools (declarative)

The pre-certified compliant agent

compliant-collector is a multilingual (Sarvam) collections agent whose flow is compliant by construction:

1. Mandatory lender-identification opening (a static node) — states the agent's name, that they are an authorised recovery agent for the named lender, the purpose, and that the call is recorded. It cannot be skipped.
2. Compliant conversation — a conversation node whose prompt enforces the no-threat / no-abuse / no-third-party tone rules and answers borrower-rights questions from rbi-rules.
3. Escalate to human — a transfer node that hands off to your compliance desk the moment the borrower disputes the debt, alleges harassment, or asks for a supervisor.

Dashboard pages

• Collection Call Compliance Monitor (dashboard) — audit-coverage %, full-population compliance rate, failed-check count, open alerts, a violations-by-rule breakdown, and a human-vs-AI compliance split. Also powers the platform dashboard_widget slot in compact form.
• Audit queue — every call with its source (AI/human), verdict and score; Run audit scores all unscored calls via App-AI; per-row Call now places a fresh compliant call to the borrower.
• Violations — open alerts ranked by severity, with the offending transcript snippet, and one-click Resolve.
• Setup — one-click Create compliant agent, settings wiring, and the call-ingestion instructions.
• Call compliance panel — the call_detail_panel slot: on any call in the dashboard it shows that call's five-rule verdict, score, and an Audit this call button.

Install & post-install wiring

1. Install Collection Call Compliance Monitor into your org.
2. Open Setup → Create compliant agent (builds the bundled Sarvam agent, auto-wired to the tools + rbi-rules KB).
3. Copy the new agent's ID into the app's Compliant collections agent ID setting (compliant_agent_id), and set the Compliance escalation desk (escalation_phone) — then update the agent's transfer-node destination to the same number.
4. Confirm the calling window (calling_window_start / calling_window_end, default 08:00 / 19:00).
5. Feed calls in:
   • AI agent calls flow in automatically via the call.analyzed event rule (no setup).
   • Human collector calls — point your dialer/CRM's call completed webhook at the app's call_log inbound hook (map call_id, phone, source: "human", collector, transcript, call_hour). The URL carries ?token=<install signing secret> for auth.
6. Open the Audit queue and press Run audit.

Settings

Bulk dialing a remediation list

The real-time path (webhook/event → workflow → call) covers new calls. To dial a remediation or apology-callback list in bulk, use the app-key campaign API:

POST /api/app-campaigns
{ "agentId": "<compliant agent id>",
  "targetQuery": { "scored": true },        // over the `call_record` object
  "result": { ... write-back ... } }

Monetization

Collection Call Compliance Monitor is blacklist insurance, priced on what it protects:

• Per audited call — a small fee for each call scored across all five rules (the coverage the RBI feedback loop expects you to have).
• Per monitored agent / collector — a monthly seat for each AI agent or human collector under 100% audit, because one un-audited breach on any of them is the risk being insured.

Open violations and their severity are surfaced on the dashboard so the ROI — breaches caught before they reach the RBI feedback loop — is obvious.

Changelog

1.1.0 — 2026-07-12

• Renamed to Collection Call Compliance Monitor (business-friendly) with a marketplace icon.
• Added real UI screenshots (dashboard + key pages).
• Verified end-to-end loop closure: trigger → workflow → call → agent tools → write-back → dashboard.

1.0.0

• Initial release.
• Data model: call_record (call_id, agent_id, phone, direction, source ai/human, collector, transcript, call_hour, score, scored, created_at) with unaudited / human_calls / ai_calls saved views; compliance_check (call_id, rule, verdict, snippet, score) with a failures view; violation_alert (call_id, rule, severity, resolved) with an open view.
• Agent tools (declarative): log_check (create a compliance check), raise_alert (open a violation alert), resolve_alert (close all alerts for a call).
• Pre-certified compliant agent: compliant-collector — a multilingual (Sarvam) collections flow: a mandatory lender-identification + recording-disclosure opening node → a compliant conversation node with no-threat / no-abuse / no-third-party tone rules → a transfer node that escalates disputes to a human compliance desk. Auto-wired to the tools and the rbi-rules knowledge base.
• Knowledge base: rbi-rules — the five RBI Fair-Practices rules (calling window, mandatory disclosure, no abuse, no threats, no third-party disclosure) written as scorable rubrics with pass/fail criteria and example breach snippets.
• App-AI scoring: the Audit page's Run audit scores every unscored call's transcript against all five rules via getTelenow().ai.llm (billed via the ai:llm scope), writing a compliance check per rule and a violation alert on each fail — guarded with a deterministic rubric fallback so the demo works from seeded data.
• Automation: call.analyzed event rule upserts a call record for each analysed AI-agent call; call_log inbound webhook ingests human-collector call logs into the same object; escalate_violation workflow (delay ~2 min → outbound call to {{settings.escalation_phone}} via {{settings.compliant_agent_id}}) rings the compliance desk on every new violation.
• Dashboard UI: audit-coverage %, full-population compliance rate, failed-check and open-alert counts, a violations-by-rule breakdown, and a human-vs-AI compliance split; also renders into the dashboard_widget slot in compact form.
• Audit queue: source / verdict / score per call, source & status filters, one-scan Run audit, and per-call Call now (dials the compliant agent with the borrower's context variables).
• Violations page: open alerts ranked by severity with the offending transcript snippet and one-click resolve.
• Call compliance panel: a call_detail_panel slot extension showing the selected call's five-rule verdict, score, and an audit action.
• Setup page: one-click Create compliant agent, settings + escalation wiring, and call-ingestion instructions.
• Self-seed: a fresh install auto-loads six realistic collections calls (three human, three AI; three pre-scored with checks + alerts, three left for the first audit run) so the dashboard is populated on first open.