Reach and right-party contact
How many dialed numbers are answered, and how many of those who answer are the actual debtor. These two rates show data quality before they say anything about conversation quality.
DIGITAL EMPLOYEE · COLLECTIONS
TAHSİLDAR: Run Collection Calls End to End with AI
TAHSİLDAR is a voice AI agent built for the work no team enjoys: reaching overdue customers, verifying who they are, explaining what they owe, and agreeing on how it gets paid. It pulls the debtor list from your collection system, places the call, and handles the conversation the way a trained representative would.
Every interaction is written back to your systems the moment it ends, outcome, promise-to-pay, and payment plan included. Collections stops being a manual workload and becomes a process you can run at institutional scale, with the same script and the same rigor on every single call.
Payment plan agreedcalm tone
Target Impact
In TAHSİLDAR projects, success is measured by the outcome of the conversation, not the number of calls. These are the areas each institution tracks with its own data.
Promise-to-pay and promises kept
How many conversations end with a payment promise, and how many of those promises turn into actual payments. Read separately, each one misleads.
Process consistency
Identity verification, the legal notice and the debt disclosure are completed in full on every call. When an audit asks "was this step applied on every call?", the goal is an answer with no exceptions.
Operational capacity
How many agents' workload the automated volume equals. Campaigns and month-end peaks scale without another hiring cycle.
See how these indicators are calculated on our Success Metrics page
NoteNumerical results depend on the institution's portfolio, delinquency profile and current process. We don't commit to rates before a pilot measured on the institution's own data.
CALL ATLAS
The Decision Network Behind a Single Collection Call
A collection call does not follow a straight script. Every answer opens a new path: identity verification, a third party, a promise to pay, a dispute, a restructuring request or a death notice. At every step TAHSİLDAR chooses the path that follows the institution's rules and writes the outcome back to the collection system.
- Rule Gate
- AI Decision
- Rule and AI
- Call Outcome
- Protection Rule
61 Decision Points111 Transitions20 Possible Outcomes
The map plays real call scenarios in turn. Drag to rotate, hover or tap a point to reveal its connections.
The map is a simplified version of the decision structure of a live collections project. For every institution, the decision points are rebuilt around its own rules and compliance framework.
Decision points and transitions in the call atlas
- Identity Verification (Rule Gate): The call does not move forward until the person on the line is confirmed as the debtor. No debt, product or call purpose is shared before that. "Yes, it's me" → Explicit Confirmation; "I'm the spouse / sibling" → Third Party; "No, that's not me" → Wrong Person; "You've got the wrong number" → Wrong Number; "I'm not available" → Not Available; "I don't want to talk" → Temporary Refusal; "No confirmation" → Verification Failed; "They passed away" → Death Notice; "It's me, but I'm at a funeral" → At a Funeral; "Don't call again" → Do Not Call Request; "They're on military service / in hospital" → Unreachable
- Explicit Confirmation (Rule Gate): The person clearly confirms who they are. The call moves on to the debt notice. "Explicit confirmation" → Debt Notice
- Third Party (Rule Gate): A relative answers. The debt is not shared; one question clarifies who owns the number. "Yes, it's their number" → Reachable Time; "It's not theirs" → Wrong Number Record; "Unclear answer" → Leave a Note; "One second, I'll hand over" → Phone Handover
- Wrong Person (Rule Gate): The person is not the debtor. Small name differences caused by speech recognition are not counted as a wrong person. "Wrong number" → Wrong Number Record
- Wrong Number (Rule Gate): The number belongs to someone else. The record is closed accordingly. "Wrong number" → Wrong Number Record
- Not Available (Rule Gate): The person cannot talk right now. The call moves to the callback flow without sharing the debt. "Call me later" → Callback
- Temporary Refusal (Rule Gate): The person does not want to talk. One polite attempt is made, with no pressure. "Okay / go ahead" → Identity Verification; "No / I don't want to" → End of Call; "I don't want this again" → Do Not Call Request
- Verification Failed (Call Outcome): No clear confirmation comes to the last verification question either. The call ends without sharing any information.
- Wrong Number Record (Protection Rule): The number is permanently flagged and the same person is not called again.
- End of Call (Call Outcome): The refusal stands after the attempt. The call ends without any debt information.
- Unreachable (Rule Gate): The debtor is in hospital, on military service or abroad. A short, fitting note is left. "Leave a note" → Leave a Note
- Phone Handover (Rule Gate): A relative hands the phone to the debtor. The agent waits and resumes from verification. "It's me" → Explicit Confirmation
- Reachable Time (Rule Gate): The agent asks when the debtor can be reached. The subject of the time is always the debtor. "Call at 11:55" → Time Gate; "I don't know / I'll tell them" → Leave a Note; "You can tell me" → Data Protection Closing
- Leave a Note (Call Outcome): The debtor cannot be reached. A fixed note is left for a callback and the call ends.
- Data Protection Closing (Call Outcome): A third party asks for information. Under data protection rules, the call ends without sharing it.
- Debt Notice (Rule Gate): The amount due and the overdue status are shared from system data, handling single and multiple products separately. "Yes / Tell me the details" → Product Details; "I don't need details" → Promise-to-Pay Request; "Tomorrow / Friday" → Date Given; "I'll pay" → No Date Given; "I paid / I deposited it" → Already Paid; "I can't pay / I have no money" → Refusal to Pay; "Call me later" → Call Me Later; "I have 500 lira" → Specific Amount; "Question about the debt" → Approved Answers; "They passed away" → Death Record; "Connect me to a person" → Live Agent; "Weather / politics" → Out of Scope; "I want installments" → Restructuring Request; "I don't owe anything" → Debt Dispute; "I'm ill / I lost someone" → Empathy
- Product Details (AI Decision): A customer with several products can hear each product's balance on request. "Details shared" → Promise-to-Pay Request
- Promise-to-Pay Request (AI Decision): The customer is asked for permission to record a payment date. A plain yes does not count as a date. "Tomorrow / Friday" → Date Given; "I'll pay" → No Date Given; "I paid / I deposited it" → Already Paid; "I can't pay / I have no money" → Refusal to Pay; "Call me later" → Call Me Later; "I have 500 lira" → Specific Amount; "Question about the debt" → Approved Answers
- Date Given (AI Decision): The customer names a day. It goes to the date validation gate. "I'll pay tomorrow / Friday / today" → Date Validation
- No Date Given (AI Decision): The customer says they will pay but gives no day. Nothing is recorded until the date is clear. "When I get my salary" → Payday; "Customer states a date" → Date Validation
- Payday (AI Decision): The customer will pay after receiving their salary. The payday is asked, not assumed. "Specific date for payday" → Date Validation
- Already Paid (Rule Gate): The customer says the payment is made. The statement is taken and no new date is asked for the paid product. "All products paid" → Payment Statement; "Products remain" → Single Negotiation Round
- Payment Statement (Call Outcome): A payment statement is taken for all products and the call ends.
- Refusal to Pay (AI Decision): The customer says they cannot pay. Restructuring is not offered unprompted. "I can't pay" → Single Negotiation Round
- Call Me Later (AI Decision): The customer wants to talk another time. The call moves to the callback flow. "Call me later" → Callback
- Specific Amount (AI Decision): The customer states the amount they have. Which product it covers is calculated in the system, not compared by hand. "I can pay today" → Date Validation
- Single Negotiation Round (AI Decision): In financial hardship, one persuasion round is made per call. No second offer follows a refusal. "Customer gives a date" → Date Validation; "I can't pay (after negotiation)" → Closing; "Call me tomorrow" → Callback; "My spouse makes the payments" → Someone Else Pays; "They passed away" → Death Record; "Connect me to a person" → Live Agent; "Weather / politics" → Out of Scope; "I want installments" → Restructuring Request; "I don't owe anything" → Debt Dispute; "I'm ill / I lost someone" → Empathy; "Don't call again" → Do Not Call Request; "Abuse" → Abuse Warning
- Date Validation (Rule Gate): The date is checked against the institution's calendar: past dates, weekends, holidays or beyond the allowed limit. "Past, holiday or out of range" → Date Correction; "Date valid" → Explicit Confirmation Round
- Date Correction (Rule Gate): For an invalid date, the customer hears why it cannot be accepted and is asked for a new one. "Another date" → Date Validation; "Confirms the shifted business day" → Explicit Confirmation Round
- Explicit Confirmation Round (AI Decision): All product dates are read back in one sentence and explicit confirmation is awaited in a separate turn. "I'm listening" → Clarify Confirmation; "Yes / I confirm" → Record Gate; "I don't confirm / I can't pay" → Closing; "They passed away" → Death Record; "Connect me to a person" → Live Agent; "Weather / politics" → Out of Scope; "I want installments" → Restructuring Request; "I don't owe anything" → Debt Dispute; "I'm ill / I lost someone" → Empathy
- Clarify Confirmation (AI Decision): "Uh-huh" or "I'm listening" is not confirmation. The question is asked once more, more clearly. "Yes / I confirm" → Record Gate; "No confirmation" → Closing
- Record Gate (Rule Gate): With a validated date and explicit confirmation, all products are written to the collection system in one go. "Record rejected" → Record Correction; "Record saved" → Closing
- Record Correction (AI Decision): If the system rejects the record, nothing is saved halfway. A new date is validated and the record is retried in full. "New payment date" → Date Validation
- Closing (Rule and AI): The debt summary, the promise taken and the consequences of delay are delivered with the institution's approved closing text. "Promise recorded" → Promise to Pay Taken; "No promise" → No Promise
- Promise to Pay Taken (Call Outcome): A promise counts as taken only after it is successfully recorded in the system.
- No Promise (Call Outcome): There is an explicit refusal or no clear confirmation. The standard closing is read without recording.
- Callback (AI Decision): The customer is asked for a suitable time. Nothing is recorded in the turn where the time is asked. "Around three" → Time Gate; "Not today (second refusal)" → Redirect
- Time Gate (Rule Gate): The time is taken exactly as said and checked against the institution's calling hours. No rounding. "Time accepted" → Callback Scheduled; "Time out of range" → Time Correction; "Record not created" → Record Error
- Time Correction (Rule Gate): If the time is outside calling hours, the customer is asked for a time within the window. "New time" → Time Gate; "No better time" → Leave a Note
- Callback Scheduled (Call Outcome): The callback time is written to the system. It does not count as a promise to pay.
- Redirect (Call Outcome): The customer declines to give a time twice. They are directed to the institution's contact channel.
- Record Error (Call Outcome): The callback could not be created in the system. The customer is not told that it was.
- Death Notice (Rule Gate): The person who answers reports that the debtor has passed away. Condolences are offered and the relationship is asked. "I'm the spouse / neighbor" → Death Notice Closing
- Death Record (Rule Gate): A death notice at any point in the call stops the payment conversation immediately. "I'm the spouse / neighbor" → Death Notice Closing
- Death Notice Closing (Protection Rule): The caller is directed to the right channel based on their relationship, and the number is permanently flagged.
- At a Funeral (Rule and AI): The customer says they are at a funeral. The debt is not read and a suitable time is asked. "Gives a time" → Time Gate; "Not now / don't call" → Funeral Closing; "I'm his mother, we're at the funeral" → Third Party
- Funeral Closing (Call Outcome): The customer gives no time. The call ends without any late-payment warning.
- Do Not Call Request (Protection Rule): The customer does not want to be called again. No persuasion is attempted and the number is permanently flagged.
- Someone Else Pays (Rule Gate): The customer says someone else makes the payments. They are asked for a date they could pay themselves. "Repeats without a date" → Someone Else Pays; "Gives a date" → Date Validation; "No date given" → Payer Statement Closing
- Payer Statement Closing (Call Outcome): No date comes. The situation is recorded and the call ends with a closing read once.
- Approved Answers (AI Decision): Questions about interest, payment channels and similar topics get the institution's approved answers, then the flow resumes.
- Out of Scope (AI Decision): Topics unrelated to collections are politely declined. "Second off-topic request" → Out of Scope Closing
- Out of Scope Closing (Call Outcome): If the off-topic insistence repeats a second time, the call ends.
- Live Agent (Call Outcome): The customer asks for a human agent. The request is recorded and the customer is told an agent will call.
- Empathy (AI Decision): In cases like illness, loss or a new job, a short empathetic sentence is added to the main answer.
- Restructuring Request (Rule Gate): An installment request is handled only if the customer asks. The agent does not offer restructuring. "Yes, pass it on" → Restructuring Record; "They passed away" → Death Record
- Restructuring Record (Call Outcome): With the customer's consent, the request is passed to the relevant team.
- Debt Dispute (AI Decision): The customer says the debt is not theirs. The dispute is heard and recorded. "This debt isn't mine (2nd dispute)" → Dispute Record
- Dispute Record (Call Outcome): If the dispute is repeated, it is passed to the relevant team and the call ends.
- Abuse Warning (AI Decision): On genuine abuse, a warning is given first. "Second genuine abuse" → Abuse Closing
- Abuse Closing (Call Outcome): If the abuse continues, the call is ended.
WHAT TAHSİLDAR DOES ON A CALL
One agent, the full collection conversation
TAHSİLDAR mirrors the exact process a human collector follows, list to outcome, and decides what to say at each step based on how the customer responds.
List integration
Pulls the debtor list straight from your collection system and queues the calls. No manual upload, no separate dialer to feed.
Voice identity verification
Confirms it is speaking to the right person through a security step before any debt detail is shared, exactly as your compliance rules require.
Debt notification
States the outstanding amount, due date, interest, and available payment channels clearly and consistently. The same accurate message every time.
Interaction & payment planning
Listens to the customer's situation, answers questions, and where appropriate negotiates and structures a payment plan within your institution's rules.
Instant recording & write-back
Logs the outcome and updates the collection system automatically as soon as the call ends, so every promise and plan is trackable.
Run at institutional scale
Handles the full overdue portfolio with the same discipline on call one and call ten thousand. No fatigue, no drift, no inconsistent scripts.
WHAT SETS TAHSİLDAR APART
Before, during and after the call
The quality of a collection call is not decided in the moment of conversation alone. TAHSİLDAR is tested against difficult scenarios before going live, handles real field conditions during the call and leaves a measurable trace after every call.
01BEFORE THE CALL
02DURING THE CALL
03AFTER THE CALL
BUILT FOR FINANCE-GRADE COLLECTIONS
Designed around your rules, your systems, and your regulators
TAHSİLDAR was built with institutional needs and regulatory sensitivity in mind, on the enterprise AI infrastructure CBOT has refined across banking, telecom, retail, and public-sector deployments.
01
Technological infrastructure
Runs on CBOT's enterprise voice stack with IVR, SIP, and MRCP integrations, CBOT's own STT/TTS engines, and a choice of LLMs, for a natural, reliable conversation in cloud or on-premise environments.
02
Integration & system compatibility
Connects directly to your existing collection and operations systems and takes the process over end to end. From identity check to payment planning, every step follows your corporate rules and is recorded as it happens.
03
Deployment models
Deploy on-premise for maximum data control or in the cloud for fast rollout and automatic scaling. Either way, TAHSİLDAR slots into your environment without disrupting existing infrastructure.
04
TAHSİLDAR as a Service
Offered through CBOT's AaaS model with no hardware or GPU investment. Infrastructure, maintenance, and model updates are handled by CBOT; you pay for what you use and the system scales as volumes grow.
- BACKED BY CBOT'S ENTERPRISE AI TRACK RECORD AND 40+ INTERNATIONALLY RECOGNIZED AWARDS ACROSS BANKING, PUBLIC SERVICES, RETAIL, AND INSURANCE.
Questions teams ask before they deploy TAHSİLDAR
Does TAHSİLDAR really handle the whole call without a human?
Yes. It retrieves the debtor list, places the call, verifies identity, notifies the customer of the debt, negotiates a payment plan where appropriate, and writes the outcome back to your collection system. End to end, no human intervention required.
How does it verify it is talking to the right person?
TAHSİLDAR runs a voice-based identity verification step before sharing any debt detail. The verification logic follows your institution's security and compliance rules.
Will it fit our existing collection system?
It integrates directly with your current collection and operations systems through IVR, SIP, and MRCP, takes the process over, and records every interaction back into those systems instantly so nothing lives in a silo.
Can we keep everything on-premise?
Yes. TAHSİLDAR can be deployed on-premise for maximum data control or in the cloud for rapid rollout and automatic scaling. Your security and infrastructure needs decide which.
Do we need to buy GPUs or run our own models?
No. Through CBOT's As-a-Service model there is no hardware or GPU investment. CBOT manages the infrastructure, maintenance, and model updates, and you pay only for what you use.
Is it built for regulated collection conversations?
TAHSİLDAR was designed with institutional needs and regulatory sensitivities in mind, using CBOT's own STT/TTS technologies and enterprise voice AI architecture trusted across the finance sector.
How does TAHSİLDAR's voice side work?
TAHSİLDAR runs on CBOT's own speech technologies: the list comes in, the call is placed, identity is verified, the debt is explained, a plan is agreed and the outcome is written back to your system. See the full flow on our AI Voice Agent for Debt Collection page.
How is the collection process automated end to end?
From list integration to writing the outcome back into your collection system, every step runs automatically within the institution's rules. We cover the scope of automation on our Automated Collections System page.
What approach does TAHSİLDAR take with overdue debt?
It verifies identity first, then explains the debt clearly and sets up a flexible payment plan within the limits the institution allows. It does not read a fixed script. More on our Overdue Debt Recovery page.
How is TAHSİLDAR different from digital collection systems?
Digital collection makes paying easier; TAHSİLDAR talks to the overdue customer and manages the debt. The side that starts the process is different. We compare the two on our Collections vs Digital Collections page.
How does TAHSİLDAR comply with collection rules and data protection law?
Calling hours, the legal notice, a clear purpose for data processing and role-based access are part of the design. We cover the compliance framework at a general level on our Compliance and Regulation page.
What is TAHSİLDAR's approach to AI governance?
It is designed around risk assessment, human oversight, explainability and model lifecycle principles, in line with the ISO/IEC 42001 framework. Details on our ISO 42001 and AI Governance page.
How does TAHSİLDAR personalize written collection messages?
Messages are not a fixed template: the institution's rules decide which tone reaches which customer at which stage, while the legal wording stays fixed. See sample drafts on our Collection Message Examples page.
Does TAHSİLDAR work on other channels, such as SMS and WhatsApp?
Its strength is the voice call; written channels can be set up with the same rules through CBOT's messaging AI Agents. We describe the scope plainly on our Omnichannel Collections page.
How does TAHSİLDAR work in banking?
It integrates with the bank's collection and operations systems, and its rules come from the bank's compliance framework. The sector-specific approach is on our Collections in Banking page.
How is TAHSİLDAR's success measured?
Not by the number of calls, but by indicators such as promise-to-pay rate, promises kept, right-party contact rate and the attribution window. The full methodology is on our Success Metrics page.
READY WHEN YOU ARE
Turn collections from a manual grind into a process you can scale.
See TAHSİLDAR handle a real overdue-payment conversation (identity check, debt notification, and payment plan) and watch the outcome land in a collection system in real time.