Root Cause Analysis Methodology

Contact centers serve as vital hubs for customer interactions, offering support, handling inquiries, and resolving issues. Analyzing these conversations can provide valuable insights into customer needs, pain points, and overall satisfaction levels. However, the sheer volume of interactions often overwhelms traditional analytical approaches, and may only result in surface-level insights, or an approach that is not scaleable or repeatable.

To address this challenge, Call Journey CI has leveraged AI-powered models capable of not only analyzing customer conversations but also identifying root causes of contact and dissatisfaction. This article gives an overview of the architecture and methodologies behind the solution, emphasizing its ability to classify calls into relevant groups using generative AI.

Architecture Overview

The architecture is subject to ongoing adaptations, but generally comprises the following components:

Automatic Speech Recognition & Transcription: Voice recordings are processed through our ASR system to convert speech into text transcripts for further analysis.
Data Ingestion: Conversational data, including voice recordings and text transcripts, undergo preprocessing steps such as cleaning, PCI redaction (or PII redaction if applicable) and anonymization to prepare for analysis. Conversational data also includes a range of metadata detailed here.
Natural Language Processing (NLP): NLP techniques are applied for sentiment and emotion analysis to enrich the conversational data
Prompt Engineering: Specific prompts guide the generative AI model in accurately summarizing and categorizing customer conversations.
1. Conversation Summary: Specific prompting guides the model in generating a concise conversation summary that includes all key events in the conversation and excludes redundant information. This summary acts as a vector for further analysis.
2. Root Cause Analysis: The model is prompted to analyze linguistic cues and use contextual understanding to identify and describe the underlying reasons for customer contact, dissatisfaction and repeat contact. This includes text-based outputs that operate as vectors for classification, including:
  1. Contact Driver Reason
  2. Dissatisfaction Reason
  3. Repeat Contact Reason
Generative AI & Prompt Engineering for Grouped Classification: The subsequent classification prompts are engineered on a per-industry basis. The model generates 3 tiers of applicable categories based on a stratified sample set from each industry. Each interaction is then processed against the model, and prompt engineering guides the model in assigning each interaction to a category set
Validation Techniques: Validation techniques, including cross-validation and human-in-the-loop validation, ensure the accuracy and reliability of the AI model's classifications. The text-based output quality is especially critical as it operates upstream of subsequent classification and category analysis.
1. Text-Based Outputs: Text outputs such as the conversation summary are human-validated against four key quality criteria on a scale of 1-5, with a minimum pass threshold of 80%:
  1. Relevance - the importance of summary content with respect to the source
  2. Consistency - the faithfulness of the summary content to the source
  3. Fluency - the linguistic quality of individual sentences in the summary
  4. Coherence - the quality of the collective structure of the sentences in the summary
2. Classifiers & Categories: Classifiers and categories are also human-validated with a minimum pass threshold of 80% accuracy. The assessment is undertaken on a stratified sample set of records in that particular industry.

Note that this architecture is subject to ongoing alteration for the purposes of improvement. With rapid advancements in LLM technology and prompt engineering approaches, Call Journey CI may alter the model or prompts that we use to ensure the best outcome for our customers.

Benefits and Applications

The AI-powered model for root cause analysis and call classification offers several benefits for contact centers:

Improved Operational Efficiency: By identifying the root causes of customer contact, contact centers can streamline processes, address recurring issues, and reduce unnecessary interactions, leading to enhanced operational efficiency.
Enhanced Customer Experience: Understanding the reasons behind customer dissatisfaction allows contact centers to prioritize and resolve issues promptly, leading to higher levels of customer satisfaction and loyalty.
Data-Driven Decision Making: Insights gained from analyzing customer conversations empower contact center managers to make data-driven decisions regarding resource allocation, training needs, and service improvements.
Scalability and Adaptability: AI-powered models can handle large volumes of data and adapt to changing customer behaviors and preferences, making them scalable and adaptable solutions for contact centers of all sizes.