For subsidiaries that provide lending and credit services such as AutoX and CardX, maintaining fair and transparent customer interactions is not only good practice but also a regulatory requirement. The Bank of Thailand requires financial institutions to operate according to strict Market Conduct standards, ensuring that customers receive clear information, fair treatment, and responsible financial guidance.
However, ensuring that tens of thousands of interactions between agents and customers meet these standards is a significant operational challenge.
The traditional approach relies on mystery shoppers. External auditors pose as real customers and interact with agents to evaluate whether services are delivered according to regulatory guidelines. While this method provides useful insights, it comes with several major limitations.
- Limited coverage
Mystery shopper programs can only evaluate a small sample of interactions. As a result, both the institution and regulators receive only a partial view of whether market conduct standards are consistently followed across the organization.
- High operational cost
Running these programs requires significant time, effort, and coordination. Recruiting auditors, conducting interactions, and reviewing results is resource intensive, making it difficult to scale the process effectively.
- Delayed feedback loops
Mystery shopper reports and feedback typically take time to circulate, often a few days or even one to two weeks, before reaching the branches or agents evaluated. By the time agents receive feedback, the service has long since been delivered, and in many cases, agents no longer remember the specific details of the customer interaction being reviewed. This delay significantly reduces the effectiveness of coaching and corrective action.
Together, these constraints make it difficult to achieve comprehensive and timely oversight of customer interactions.
The Approach:
AI Powered Market Conduct Monitoring
To address these challenges, SCBX explored how large language models and AI driven analysis could automate the monitoring of customer interactions while maintaining regulatory standards.
Instead of evaluating only a limited number of interactions through manual audits, the system analyzes conversation transcripts and automatically evaluates whether agents follow required conduct guidelines.
This approach enables institutions to monitor interactions continuously and at scale, providing a far more complete picture of customer treatment, and delivering feedback fast enough to be genuinely useful.
Example Applications
AutoX: Market Conduct Detection
For AutoX, the system focuses on ensuring that financial products such as loans and insurance are offered in a responsible and transparent manner.
When an agent interacts with a customer, the conversation is captured through the agent’s smart device. The system then uses Automatic Speech Recognition (ASR) technology to transcribe the voice conversation into text in real time. Once transcribed, the conversation is analyzed by an AI model that checks whether the interaction follows the Market Conduct standards required by the Bank of Thailand.
The system evaluates whether agents clearly explain product terms, disclose risks appropriately, and avoid coercive or misleading sales tactics.
Unlike traditional manual audits that review only a small sample of calls, the AI system can monitor every recorded interaction, significantly improving compliance coverage and reducing regulatory risk.
Near real-time feedback for agents
Another key advantage, and arguably one of the most impactful, is the dramatic reduction in feedback time. In traditional mystery shopper processes, it can take days or weeks for a compliance issue to surface, travel up through managerial reviews, and eventually reach the agent. By that point, the agent may have forgotten the details of the interaction, making the feedback far less actionable.
Our current proof of concept demonstrates an AI response time of just a few minutes, averaging 5 to 8 minutes per interaction, with the potential to be significantly faster as execution capacity scales. Results are sent directly to the service agent’s email, allowing them to review their performance while the interaction is still fresh in their mind.
This shift from weeks to minutes transforms feedback from a retrospective audit into a real-time learning loop. Agents can recognize and correct mistakes almost immediately, which makes coaching more effective and behavioral improvement more sustainable over time.
CardX: Quality Control in Debt Collection
For CardX, the focus is ensuring that debt collection interactions remain respectful, professional, and compliant with legal standards.
The system analyzes conversation content and tone to determine whether collectors maintain appropriate professionalism and provide required explanations to customers. If required disclosures or explanations are missing, the system can flag these issues and alert supervisors.
By automatically generating performance reports and behavioral scores, management gains better visibility into service quality and can provide more targeted coaching to improve customer experience.
From Experimentation to Organizational Impact
These solutions demonstrate how AI can transform compliance monitoring from a manual sampling process into a scalable oversight system.
By enabling continuous monitoring across all customer interactions, the approach significantly improves regulatory visibility while reducing operational costs. Just as importantly, the near real-time feedback loop ensures that insights translate into actual behavior change at the agent level, rather than being lost to the delays of traditional auditing cycles.
The system provides both institutions and regulators with stronger assurance that market conduct standards are consistently upheld.
This innovation, which has been patented, represents a step toward more transparent, accountable, and customer centered financial services across the SCBX ecosystem.
