AI Voice Agent for Debt Collection

A collections call is one of the most sensitive conversations an institution has with its customer. TAHSİLDAR runs that conversation by voice: it pulls the debtor list from your collections system, places the call, verifies identity, explains the balance, sets up a payment plan where the rules allow it, and writes the outcome straight back into your system.

The technical basis of the voice layer

CBOT's own speech technologies

TAHSİLDAR's voice is not a service rented from elsewhere. The engines that turn speech into text and text back into speech are developed by CBOT. That has three practical consequences: the pronunciation and inflection of a language such as Turkish can be worked on directly, voice data does not have to leave the institution's boundaries, and latency can be measured and improved on the institution's side.

At the understanding layer there is no lock-in to a single model. Different LLMs can be used depending on the nature of the conversation and the institution's preference. There is a fuller account on our Voice AI page.

Natural sentences instead of keypad menus

Classic interactive voice systems bend the customer to the menu: press one to pay. TAHSİLDAR does the opposite. When the customer says "this month is difficult, I can pay on the twentieth", the intent is understood and the conversation continues from there. No menu, no waiting, no repetition.

The call, step by step

How a collections conversation runs from start to finish

The list is pulled. The debtors to be called are drawn from your collections system. There is no intermediate step of uploading a file or copying a spreadsheet.

The call starts. The queue is built automatically; who is called, when, and how many times follows the institution's rules.

Identity is verified. Before any detail of the debt is shared, the agent confirms by voice that it is speaking to the right person. The verification logic comes from the institution's security and compliance rules.

The balance is explained. Amount, due date, delinquency status and payment channels are stated clearly.

A plan is set up. Where the customer's situation allows and the institution permits it, a payment plan is created.

The outcome is written back. The result of the call is posted to the collections system immediately. No separate transfer, end of day file or manual entry.

Why the first ten seconds decide the call

Building trust on an outbound call

On an inbound call the customer is already looking for you. Outbound is the reverse: the person who picks up does not know what to expect and their default posture is guarded. That is why the opening seconds decide the fate of everything that follows.

TAHSİLDAR makes three things clear at the opening: who is calling, why, and who it needs to speak to. Every second left ambiguous is a call that ends. This is also where the voice channel has an advantage: tone, pace and clarity carry a trust signal that a written message cannot.

We describe how a voice AI agent that follows the institution's rules and writes results back is positioned on the TAHSİLDAR overview page.

How the voice agent and the human team work together

The job of a voice AI agent is not to replace the human team but to move the team's time to where it counts. High volume, repetitive conversations with predictable outcomes run automatically. When there is a dispute, a special case or a request that falls outside the boundaries the institution has defined, the conversation is handed to a person.

That split is designed in from the start: the conditions for handover are the institution's rule, not the model's decision. Context travels with the handover, so the customer does not have to explain the same thing a second time.