AI-Powered Dunning

BillHive’s AI Dunning Engine replaces one-size-fits-all retry schedules with personalized, adaptive recovery strategies. Instead of following the same timeline for every customer, the system predicts the best recovery action for each individual case — when to retry, how to communicate, and what offer to make.

Why AI Dunning

Traditional rule-based dunning applies the same schedule to every customer. AI-driven dunning treats each failed payment as a unique recovery opportunity.

AI Components

BillHive’s AI dunning system consists of 10 interconnected components that work together to maximize payment recovery.

1. Payment Recovery Prediction

When a payment fails, the AI model predicts the probability that the payment will be recovered. This score drives all downstream decisions.

Input features:

Output:

{
  "recovery_probability": 0.82,
  "confidence": 0.91,
  "recommended_strategy": "retry_soon",
  "estimated_recovery_days": 3
}

Decision matrix:

2. Smart Retry Scheduler

Instead of retrying on fixed days, the AI predicts the optimal retry window for each customer based on historical patterns.

How it works:

• Analyzes past payment success patterns by day of week and time of day
• Detects salary cycle patterns (e.g., payments succeed around the 1st and 15th)
• Identifies bank-specific decline windows (e.g., some banks decline more at night)
• Avoids retrying during known low-success windows

Example optimizations:

Impact: Smart retry timing alone can improve recovery rates by 10–20% compared to fixed schedules.

3. AI Communication Generator

Instead of static email templates, the AI generates personalized dunning messages tailored to the customer context.

Personalization inputs:

Example — Friendly tone for a startup customer:

Hi John,

Looks like your payment of $299 didn’t go through — this sometimes happens when cards hit their limit.

You can update your card here: [Update Payment Method]

We’ll retry automatically tomorrow. No action needed if your card is fine — it might just be a temporary hold.

Example — Formal tone for an enterprise customer:

Dear Finance Team,

Invoice #INV-2024-1847 for $12,500.00 was not processed successfully. The payment gateway reported: card expired.

Please update your payment method at your earliest convenience: [Update Payment Method]

Your subscription services remain active during the grace period. If you need to arrange alternative payment terms, please contact your account manager.

4. Customer Behavior Scoring

AI continuously scores each customer’s likelihood to churn vs recover, using product engagement signals alongside payment data.

Scoring signals:

Risk-based dunning actions:

5. Channel Optimizer

AI determines the best communication channel for each customer based on historical engagement data.

Channel selection logic:

Learning loop: The system tracks open rates, click rates, and payment conversions per channel per customer segment, and shifts channel allocation toward higher-performing channels over time.

6. AI Negotiation Agent

An automated conversational agent that handles real-time payment recovery conversations with customers.

Capabilities:

Guardrails:

• Maximum discount authority: configurable per tenant (e.g., up to 20% for 1 billing cycle)
• Escalation to human agent when confidence is low or customer is frustrated
• All offers are logged and tracked for revenue impact analysis
• Configurable offer policies per plan tier

7. Root Cause Analysis

AI analyzes failure patterns across all dunning cases to identify systemic issues and recommend preventive actions.

Pattern detection examples:

Actionable outputs:

• Monthly dunning health report with top failure reasons
• Gateway performance comparison (which gateway recovers better)
• Payment method recommendations (e.g., “Enable UPI Autopay to reduce failures by 15%”)
• Billing date optimization suggestions

8. Customer Segmentation

AI automatically clusters customers into segments with distinct dunning strategies, rather than applying one strategy to all.

Auto-generated segments:

Segments are dynamic — customers move between segments as their behavior changes.

9. Self-Learning Engine

The dunning system continuously improves by learning from every recovery attempt.

Feedback loop:

Payment Fails
    → AI predicts recovery strategy
    → Strategy executes (retry, email, offer)
    → Outcome recorded (recovered / not recovered)
    → Model retrained on new outcome data
    → Future predictions improve

Metrics tracked per strategy:

A/B testing framework:

• Automatically split dunning cases into control (rule-based) and treatment (AI) groups
• Compare recovery rates, time-to-recovery, and customer satisfaction
• Gradually shift traffic to winning strategies
• Configurable experiment duration and sample size

10. Predictive Churn Prevention

Goes beyond reactive dunning by predicting payment failures before they happen and taking preventive action.

Pre-dunning signals:

Architecture

System Overview

┌─────────────────────────────────────────────────────────┐
│                    Payment Fails                         │
└──────────────────────┬──────────────────────────────────┘
                       ▼
┌─────────────────────────────────────────────────────────┐
│              AI Dunning Orchestrator                     │
│  ┌─────────────┐  ┌──────────────┐  ┌───────────────┐  │
│  │  Recovery    │  │  Behavior    │  │  Segmentation │  │
│  │  Predictor   │  │  Scorer      │  │  Engine       │  │
│  └──────┬──────┘  └──────┬───────┘  └──────┬────────┘  │
│         └────────────────┼─────────────────┘            │
│                          ▼                               │
│  ┌─────────────────────────────────────────────────┐    │
│  │          Strategy Decision Engine                │    │
│  └──────────────────────┬──────────────────────────┘    │
└─────────────────────────┼───────────────────────────────┘
                          ▼
        ┌─────────────────┼─────────────────┐
        ▼                 ▼                 ▼
┌──────────────┐ ┌──────────────┐ ┌──────────────────┐
│ Smart Retry  │ │  Channel     │ │  Communication   │
│ Scheduler    │ │  Optimizer   │ │  Generator (LLM) │
└──────┬───────┘ └──────┬───────┘ └──────┬───────────┘
       │                │                │
       ▼                ▼                ▼
┌─────────────────────────────────────────────────────────┐
│              Execution Layer                             │
│  Payment Retry API  │  Notification Engine  │  Chatbot  │
└──────────────────────┬──────────────────────────────────┘
                       ▼
┌─────────────────────────────────────────────────────────┐
│              Outcome Tracking & Learning                 │
│  Recovery logged → Model retrained → Strategies updated │
└─────────────────────────────────────────────────────────┘

Integration with Existing BillHive Services

Job Queues (BullMQ)

Database Schema

dunning_ai_cases

Extends the base dunning case with AI-specific data.

dunning_ai_actions

Tracks every AI-driven action taken during a dunning case.

dunning_ai_outcomes

Records the outcome of each action for model training.

dunning_ai_model_metrics

Tracks model performance over time.

Configuration

AI dunning is configured via Settings > Dunning > AI Settings in the BillHive dashboard.

AI Dunning Settings

AI Dunning Configuration API

GET /dunning-config/ai

Retrieve the AI dunning configuration for the tenant.

Response:

{
  "enabled": false,
  "mode": "shadow",
  "llmProvider": "claude",
  "negotiation": {
    "maxDiscountPercent": 20,
    "discountDurationCycles": 1
  },
  "abTestSplit": 50,
  "churnPreventionEnabled": false,
  "enabledChannels": ["email", "sms"]
}

PUT /dunning-config/ai

Update the AI dunning configuration.

Validation rules:

Rollout Strategy

AI dunning should be rolled out gradually to minimize risk and build confidence.

Phase 1: Shadow Mode (Weeks 1–4)

• AI runs alongside rule-based dunning but does not execute actions
• AI predictions and recommendations are logged for comparison
• Compare AI-recommended retry times vs actual retry results
• No customer impact

Phase 2: Assist Mode (Weeks 5–8)

• AI recommends actions; human reviews and approves
• Start with smart retry timing only (lowest risk)
• Add AI-generated communications after validating retry timing
• Measure recovery rate delta vs rule-based

Phase 3: Full Mode — Retry & Communication (Weeks 9–16)

• AI controls retry scheduling and message generation
• A/B test with 50% AI, 50% rule-based
• Track: recovery rate, time-to-recovery, customer satisfaction
• Expand to all enabled channels

Phase 4: Full Mode — Negotiation & Prevention (Week 17+)

• Enable AI negotiation agent for self-serve interactions
• Enable predictive churn prevention
• Expand A/B test to 80% AI
• Continuous model retraining on outcome data

Metrics & Reporting

AI Dunning Dashboard

Webhook Events

Competitive Advantage

Most billing platforms still use rule-based dunning. BillHive’s AI dunning positions the platform ahead of major competitors.

Related Topics

• Dunning Overview — How BillHive dunning works today
• Configure Dunning — Rule-based dunning configuration
• Pending Features — Planned enhancements roadmap