Automated Debt Collection, Dunning & Accounts Receivable Recovery Systems: Combining SMS, WhatsApp, Email, and AI Voice

Why Manual Debt Collection Teams Fail to Scale—and Alienate Good Customers

Whether you operate a digital credit or asset-financing business, a B2B wholesale distributor extending 30-day credit terms, a fiber ISP/utility billing 25,000 households, or an insurance/property management firm, overdue Accounts Receivable (AR) directly throttles your cash flow. Yet traditional recovery methods are broken at both ends: human call-center agents cost $400–$800/month per seat, spend 70% of their day listening to unanswered ringtones, forget to follow up on Promise-to-Pay (PTP) dates, or use aggressive tones that permanently alienate loyal customers who simply forgot an invoice.

Meanwhile, blasting identical bulk SMS reminders to every overdue account quickly trains borrowers and B2B clients to ignore your messages. At devManue Conversational AI & Fintech Engineering, we build intelligent Multi-Channel Dunning & Accounts Receivable Recovery Engines that orchestrate personalized SMS, WhatsApp Interactive Payment Buttons, formal PDF statements, and polite, sub-500ms AI Voice Collection Agents—triggering live M-Pesa STK Push prompts or structured Promise-to-Pay restructurings while the customer is still on the phone.

1. Multi-Channel Dunning & AI Collections Architecture Modules & Costs

Our collections platform replaces chaotic spreadsheets and manual dialers with a deterministic, compliance-enforced state machine:

2. How the Dunning State Machine Coordinates WhatsApp, Voice AI, and Instant Suppression

Nothing damages customer trust faster than an automated collection robot calling a client at 4:00 PM to demand payment for an invoice they already paid via M-Pesa at 10:30 AM. Our dunning engine is wired directly into your PostgreSQL payment webhook ledger: the exact millisecond a full or partial payment callback arrives, any scheduled SMS, WhatsApp message, or outbound AI phone call for that account is atomically cancelled and replaced with an instant digital receipt.

When an account enters Days Past Due (DPD) 3 without responding to text notifications, our outbound AI Voice Agent places a polite phone call during approved daytime hours. If the customer says, "I can pay KES 5,000 right now and the remaining KES 7,000 on Friday the 25th," the AI agent invokes a mid-stream tool call that triggers an M-Pesa STK Push for KES 5,000 straight to their phone screen while still on the line, logs a verified Promise-to-Pay (PTP) for the 25th, and pauses further calls until Friday:

# Production AI Voice Collector Mid-Call Tool Execution: STK Push & Promise-to-Pay
from datetime import date
from decimal import Decimal
from django.db import transaction
from django.utils import timezone
from .models import OverdueReceivableAccount, PromiseToPayRecord, DunningInteractionLog
from payments.services import trigger_mpesa_stk_push

@transaction.atomic
def execute_mid_call_collection_action(
    account_id: int,
    immediate_stk_amount: Decimal,
    ptp_balance_date: date | None = None,
    call_sid: str = "",
) -> dict:
    """
    Invoked live by the WebRTC/SIP Voice AI Agent during a customer call.
    Triggers an instant M-Pesa STK Push and pauses dunning until the agreed PTP date.
    """
    acct = OverdueReceivableAccount.objects.select_for_update().get(id=account_id)
    if acct.current_balance_due <= Decimal("0.00"):
        return {"status": "ALREADY_SETTLED", "message": "Account has zero balance."}

    stk_ref = None
    if immediate_stk_amount > Decimal("0.00"):
        stk_ref = trigger_mpesa_stk_push(
            phone_e164=acct.customer_phone,
            amount=min(immediate_stk_amount, acct.current_balance_due),
            account_reference=f"INV-{acct.invoice_number}",
        )

    if ptp_balance_date:
        PromiseToPayRecord.objects.create(
            account=acct,
            promised_amount=acct.current_balance_due - immediate_stk_amount,
            promised_date=ptp_balance_date,
            recorded_via="AI_VOICE_AGENT",
            call_sid=call_sid,
        )
        acct.dunning_status = "PAUSED_ON_PTP"
        acct.next_action_at = timezone.make_aware(
            timezone.datetime.combine(ptp_balance_date, timezone.datetime.min.time())
        )
        acct.save(update_fields=["dunning_status", "next_action_at", "updated_at"])

    DunningInteractionLog.objects.create(
        account=acct,
        channel="AI_VOICE",
        outcome="STK_SENT_AND_PTP_LOGGED" if ptp_balance_date else "STK_PUSH_TRIGGERED",
        call_sid=call_sid,
    )
    return {"status": "SUCCESS", "stk_checkout_id": stk_ref, "ptp_date": str(ptp_balance_date)}

3. Regulatory Compliance Guardrails Built Into the Code

Central bank digital credit regulations and data protection laws strictly prohibit predatory collection practices—such as calling borrowers at night, harassing third-party contacts who are not formal guarantors, or placing excessive daily calls. We enforce these rules at the database kernel level: our outbound dialer queue mathematically blocks any call attempt outside 08:00–19:00 local time, caps outbound attempts at a configurable weekly limit, and logs immutable transcripts and sentiment scores for every conversation.

Want to recover more overdue revenue while cutting call-center overhead by 75%? Forecast your telephony and AI voice unit economics with our Voice AI Latency & Cost Calculator, test real-time audio responsiveness in our WebRTC Audio Latency Tester, or book a 30-Minute Collections Automation Discovery Call.

4. Frequently Asked Questions (FAQ)