Who Makes Each Collections Call: AI Voice Agent, Person or IVR?

A collections operation that puts every call through people runs out of people. One that puts every call through AI loses the customers who need a person. Most lenders end up somewhere in between, and the useful question is where to draw the lines.

Bank Neo Commerce (BNC), an Indonesian digital bank, drew them across three channels: a collections AI voice agent, human agents on predictive dialing, and one-way IVR notifications. Each covers a different part of the collection process. BNC reports 80%+ labor cost savings and 5–10% repayment improvement.

Why a Single-Channel Collections Process Struggles

BNC’s challenge, in its own case study, was handling a fast-growing user base, especially in the early stages of a project launch, while balancing operational efficiency against cost.

Growth is hard on collections. The number of accounts due each month grows with the loan book, and most of the new calls are routine reminders. If every reminder and early overdue call needs a human agent, headcount has to grow in step with the book. If the bank cuts back to automated messages only, it reaches fewer customers in a way they respond to.

The answer BNC chose was to stop treating collections as one kind of call.

How the Three Channels Split the Work

One-way IVR notifications carry simple, one-directional messages: a reminder that a payment is due, with no conversation needed.

The collections AI voice agent handles the conversations that follow a pattern. It works from a script and interaction strategy built for BNC’s customers: confirming the customer, explaining what is owed, asking when they can pay and recording the commitment. These calls come in the largest volumes and gain the most from automation.

Human agents on predictive dialing take the conversations that need judgement: disputes, hardship cases, customers who have broken earlier promises, or anyone who asks for a person. BNC’s human agents work on predictive dialing (PDS), which dials ahead automatically so agents spend their time on conversations. In AI Rudder’s AI Contact Center this is the Agent Dialing System.

The split is set by the bank’s own collection strategy, and it can change. When BNC wants a different rule for which accounts go to people, it changes the strategy, and the routing follows.

What Tailoring the Script Does

The phrase in BNC’s case study that deserves attention is “customized script and interaction strategy”. A generic reminder script treats every customer the same. A tailored one reflects the bank’s products and customers. It also changes with the stage of the debt: a friendly first reminder before the due date reads very differently from a call on day 30.

AI Rudder’s in-market teams configure the agents to each client’s call flows and rules, and the client’s team can then adjust scripts and campaigns themselves. This is where years of collections calls in Indonesia and other Southeast Asian markets show up: in how the conversation is ordered and how the agent responds when a customer pushes back. It also shapes when the agent hands over.

What BNC Reports

From BNC’s approved case study:

• 80%+ labor cost savings
• 100% delay-contact resolution
• 5–10% repayment improvement
• Increased customer experience

The repayment figure matters as much as the savings. Cost went down while more customers paid, which is the result a split like this is designed to produce: routine calls on the AI voice agent, and human agents on the accounts that need them.

Questions Lenders Ask About Mixing AI and Human Agents

Which collections calls should go to an AI voice agent first? Start with high-volume, rule-based calls, such as pre-due reminders and early overdue follow-ups, where the conversation follows a known pattern.

How does a call move from the AI voice agent to a person? Transfer conditions are part of the call flow the bank defines, for example a dispute or a request to speak to someone.

Do we need new dialing software for the human agents? AI Rudder’s AI Contact Center includes the Agent Dialing System. It supports predictive dialing, and agents can also switch to progressive or manual dialing, so AI and human channels can run on one platform.

Does AI Rudder work in Indonesia? Yes. AI Rudder started in Indonesia in 2019 and has an office there.

Next Step

Map your collections process by stage and decide which calls need a person. Get a Demo and we’ll show you how the AI voice agent, human agents and IVR would split your portfolio.