Context and Need
The existing setup used in Fibabanka’s collections operation had room for improvement in the naturalness of customer dialogue, callback processes and operational efficiency.
The need was not simply for a system that spoke fluently in a demo. Fibabanka required a voice AI solution that could conduct conversations in natural Turkish under real contact-centre conditions, accurately understand customer responses and reliably take the actions required by the collections process.
Manual follow-ups for customers who could not be reached or were unavailable during the call also created a significant workload for the operations team. Automating this effort would allow the team to focus on conversations that required human intervention and had greater potential to produce results.
Solution
Mank Voice AI was deployed on-premise within Fibabanka’s own data center. Voice data and conversation processes remained within the institution’s boundaries, while the system was integrated with the operation’s existing technology infrastructure without requiring an external internet connection.
- 01
Natural Turkish collections conversations
Collections conversations were automated through natural Turkish dialogue.
- 02
Automated callbacks
Callback triggers were created for customers who could not be reached.
- 03
Appointment management
A new call appointment was scheduled at a suitable time for customers who were unavailable during the conversation.
- 04
Controlled operational flow
Conversation outcomes were recorded, and cases requiring human intervention were routed to the relevant teams.
This reduced the operations team’s need to schedule callbacks manually and moved collections conversations to a more scalable structure.
- 01
Discovery
- 02
Success criteria
- 03
Integration
- 04
Controlled production
- 05
Continuous improvement
An Insight from the Field
One concern at the beginning of the project was that customers might end the conversation when they realized they were speaking with an AI system.
Observations from live operations showed that some customers began expressing themselves more clearly and systematically after realizing they were speaking with AI. This behavior helped the system understand customer intent more accurately and enabled the conversation to proceed in a more structured way.
Results
Following the introduction of Mank Voice AI in the live collections operation, a 40–45% promise-to-pay rate was achieved among reached customers. Promise-to-pay performance improved by more than 30% compared with the previous system.
Automated callbacks and appointment management reduced the need for manual follow-up after initial calls that did not produce a result. Instead of planning repetitive calls, the operations team could focus on conversations requiring human judgment and processes that generate greater value.
The project, covering discovery, development, security assessments, integration and a controlled transition to live operation, was completed in approximately 7–8 months. Adapted to real operational conditions together with Fibabanka’s teams, the architecture has become an infrastructure that can now be extended to new use cases.
The same technology layer is also used in Fibabanka’s loan sales operation.