A story about process automation

Challenge

An Latin American telecommunications company needed to automate the quality verification process for various types of calls made to their call center. The goal was to ensure that agents adhered to specific KPIs, such as greeting callers appropriately, responding according to set protocols, avoiding repeated requests for the same information, and communicating current promotions effectively.

The company’s call center was overwhelmed with a high volume of calls, including telemarketing, customer service, and SAEC (service activation and customer experience) calls. With only seven quality agents responsible for listening to, evaluating, and rating calls based on quality parameters, they could only evaluate 30 calls per day, representing less than 10% of the total calls. This limited sample size was insufficient for ensuring a comprehensive quality assessment.

Proposed Solution

By implementing an innovative system to address this challenge by:

1. Converting call audio to text using advanced speech-to-text technology.
2. Training machine learning models to evaluate calls based on established KPIs.

The system categorized calls into two main groups:

• Compliant Calls: Automatically rated for quality standards.
• Non-compliant Calls: Flagged for human review.

To achieve this, the system developed and utilized approximately 30 different models, each tailored to assess specific KPIs. An initial challenge was acquiring labeled data to train these models effectively.

Results

The implementation brought substantial improvements in the quality verification process:

• Increased Call Evaluation: The percentage of calls evaluated monthly rose from approximately 7% to 40%.
• Optimized Agent Focus: Automation allowed quality agents to concentrate on calls flagged as potentially non-compliant, enhancing their efficiency.
• Comprehensive KPI Evaluation: Calls that could be automatically rated were assessed 100% according to the relevant KPIs.
• Targeted Human Review: Calls with low confidence scores triggered alerts for quality agents to manually review, ensuring focused attention on problematic areas.

This automation enabled the company to direct the expertise of quality agents to calls that required detailed scrutiny, providing opportunities to establish additional KPIs or conduct further reviews as necessary. By streamlining the rating process, the company could effectively address quality issues, resulting in improved overall call center performance.

‍