Ho

Proactive and effective Collections Agent

Track: Financial Services II · Truong tu Team

Elevator Pitch Proactive and Effective Collections Agent is a predictive early-warning and personalized collections solution that helps financial institutions act before a payment is missed. It identifies high-risk customers 7–14 days before their due date, recommends the right treatment, channel, timing, and message, while keeping humans in control through approval and compliance checkpoints. The result is a more proactive, effective, and customer-friendly collections process. Inspiration Traditional collections processes are often reactive, rule-based, and one-size-fits-all. Customers are typically contacted only after missing a payment, using standard messages and broad outreach campaigns. This approach creates unnecessary manual work, increases collection costs, delays debt recovery, and can negatively affect customer experience. We saw an opportunity to use AI agents to detect risks earlier, personalize each intervention, and improve collections performance without removing human oversight. What It Does Proactive and Effective Collections Agent combines customer profiles, repayment history, outstanding balances, transactions, behavioral signals, and collection history to generate predictive early-warning signals. A Risk Agent identifies each customer’s risk level and explains the key factors behind its decision. A Treatment Agent then recommends the next-best action, including the most appropriate treatment, communication channel, and contact timing. An LLM-powered Message Agent prepares a personalized outreach proposal based on the customer’s situation. Before anything is sent, a human reviewer can approve, edit, reject, or escalate the proposal. After outreach, the solution measures outcomes such as cure rate, response rate, cost-to-collect, missed payments, and complaints. It then identifies patterns and proposes improvements to the risk and treatment matrices. These changes require human approval before being applied to the next collection cycle. How We Built It We designed Proactive and Effective Collections Agent as a modular and governed agent workflow rather than a single AI model. Risk and treatment matrices provide business rules and operational guardrails. Specialized AI agents handle risk prediction, treatment recommendations, personalized messaging, and outcome analysis. Human-in-the-loop controls are included at two critical points: before customer outreach and before proposed matrix changes go live. This ensures that sensitive decisions remain explainable, auditable, compliant, and under human control. We built the interactive prototype as a self-contained HTML presentation using HTML, CSS, Vanilla JavaScript, SVG connectors, and state-based animations. The presentation demonstrates how AI agents progressively transform a traditional collections process into a predictive and continuously improving workflow. Challenges We Ran Into The main challenge was balancing automation with explainability, compliance, and human oversight. Collections decisions can directly affect customers, so every AI recommendation must be understandable, auditable, and easy for a human reviewer to override. We also needed to simplify a complex end-to-end workflow without losing important elements such as escalation, high-risk case handling, performance measurement, and continuous learning. Other challenges included defining meaningful success metrics, managing AI operating costs, designing appropriate escalation thresholds, and ensuring that personalized outreach remains respectful and compliant. Accomplishments That We’re Proud Of We created a complete transformation journey from a reactive collections process to a proactive, governed, and continuously improving operating model. The solution covers early-warning detection, risk assessment, treatment optimization, personalized messaging, human approval, multichannel outreach, performance measurement, and controlled continuous learning. We are especially proud that Proactive and Effective Collections Agent does not position AI as a replacement for collectors. Instead, it helps teams prioritize high-risk accounts, reduce repetitive manual work, and focus human expertise on complex or sensitive cases. We also designed a roadmap that takes the solution from proof of concept to controlled pilot, production product, and eventually a reusable commercial platform. What We Learned We learned that the value of agentic AI comes from orchestration, governance, and continuous learning, not simply from adding a chatbot or LLM to an existing process. High-quality data, measurable outcomes, explainable recommendations, business guardrails, and human feedback are just as important as model performance. We also learned that AI economics should be treated as a core business KPI. Model routing, prompt optimization, caching, reusable decisions, smaller models, and appropriate escalation thresholds can reduce AI cost per cured account while maintaining strong controls. Finally, deploying the solution early can create a significant learning advantage. Real-world outcomes can reveal which variables, treatments, and AI techniques have the greatest impact on cure rate, cost-to-collect, customer response, and collector productivity. What’s Next for Proactive and Effective Collections Agent The next step is to validate the solution through a focused proof of concept using real portfolio data and clearly defined KPI baselines. We plan to test the agents in shadow mode before launching a controlled pilot with selected customer segments and products. The pilot will measure predictive lift, cure-rate improvement, cost-to-collect reduction, complaint rates, compliance performance, and collector adoption. After proving business value, we will integrate the solution with core banking systems and communication channels, introduce MLOps and LLMOps monitoring, strengthen security and audit controls, and optimize AI unit economics. In the longer term, Proactive and Effective Collections Agent could become a configurable platform or advisory offering for traditional banks, digital banks, e-wallet providers, and lending institutions. By deploying and learning early, GoTyme can build proprietary collections data, risk and treatment matrices, governance patterns, and operating know-how that become increasingly difficult to imitate and potentially create new commercial opportunities.