Council Post: Customer Service Call Assist Using Generative AI

One of the paramount challenges confronting Customer Service Leads in B2C companies is the optimization of desk call agent productivity and effectiveness in addressing customer inquiries. While routine queries are typically handled by chatbots or automated voice responders, those that escalate to desk call agents tend to be more intricate in nature. These can encompass a spectrum of issues such as payment disputes, fraud reporting, or points redemptions, necessitating a profound understanding of the product, workflow, and swift comprehension of the customer’s specific circumstances. Failure to address client queries adequately and promptly can lead to adverse outcomes such as customer attrition, low CSAT/ NPS score or even the non-receipt of payments from customers. Although Customer Service managers may conduct manual audits, these efforts often prove inadequate and belated. The crux of the challenge lies in automating these critical processes in real-time, while the call is ongoing.

Real-Time Feedback on Customer Mood

Accurate Mood Detection: One of the primary challenges in automating real-time feedback on customer mood is accurately detecting and interpreting emotional cues. While AI models can be trained to recognize certain patterns in speech and tone, accurately interpreting complex emotions like sarcasm or irony can be difficult.

Privacy Concerns: Another challenge is ensuring that customer privacy is protected while collecting data on their mood. Companies must be transparent about what data they are collecting and how it will be used to avoid violating privacy regulations.

Real-Time Feedback on Agent Self-Performance

Subjectivity: Evaluating agent performance is inherently subjective, and different supervisors may have different opinions on what constitutes “borderline aggression” or other negative behaviors. Developing an AI model that can accurately identify these behaviors in real-time is a significant challenge.

Integrating with Existing Systems: Many companies use a variety of different tools and platforms to manage customer service interactions, and integrating an AI-powered feedback system with these existing systems can be challenging.

Information Retrieval Based on Context and History

Data Quality: Retrieving relevant information based on context and past interactions requires access to high-quality data. If the data is incomplete or inaccurate, the AI model may provide incorrect or irrelevant information to agents, leading to frustration and decreased efficiency.

Complexity of Queries: Some customer queries may be highly complex, requiring the AI model to analyze multiple data sources and synthesize information from different domains. Developing an AI model that can handle these complex queries in real-time is a significant challenge.

Detecting customer emotions, like anger or sarcasm, during desk calls is crucial for providing appropriate responses. When customers are visibly angry, such as shouting on the call, agents need effective comeback options to de-escalate the situation and address their concerns. For instances where customers express frustration through sarcastic comments, it’s essential to identify these cues and respond with empathy and understanding. Accurately discerning the customer’s mood and intentions allows agents to tailor their responses, ultimately aiming to satisfy the customer and resolve their issues satisfactorily.

Instant Feedback for Self-Improvement

Providing real-time feedback to call center agents can significantly enhance their performance and customer interactions. Implementing a system that offers color-coded indicators like red, green, or yellow, along with explanations for the scores, can empower agents to make immediate improvements during calls. 

Examples of Real-Time Performance Feedback Systems

Genesys PureCloud: Genesys PureCloud includes a real-time performance monitoring feature that displays agent performance scores in various categories like call resolution time, customer sentiment analysis, and compliance with company policies. Agents receive color-coded feedback along with explanations to help them adjust their approach during calls.

Zendesk Talk: Zendesk Talk integrates real-time feedback mechanisms that show agents their performance scores based on factors like call handling time, customer feedback ratings, and resolution accuracy. Agents can view color-coded indicators and detailed reasoning to improve their performance instantly.

Enhancing Information Retrieval for Complex Queries: Strategies, Examples, and Implementation

We’ve all experienced the frustration of receiving “standard template answers” from call center agents. It’s a situation we all dislike. After all, who wouldn’t prefer answers that are tailored and contextualized to their specific questions, current situation, and past interactions?

 Examples of Gen AI tools, techniques, and service providers in the realm of call center assistant systems are:

• IBM Watson Discovery: Harnesses deep learning algorithms to deliver contextually relevant answers to complex queries.

• Oracle Digital Experience: Integrates AI-based recommendation engines to provide tailored responses to customer inquiries.

• HomeServe Virtual Agent ‘Charlie’: Manages an impressive volume of over 11,000 calls per day, offering customers answers, initiating claims processes, and efficiently routing calls to agents.

Developing Call Agent Assist systems involves navigating a myriad of challenges, spanning legal, technical, data, and ethical domains. Furthermore, the decision to build or buy such systems warrants careful consideration of factors like customization, control, cost, and expertise. Let’s delve into these aspects and explore the landscape of Call Agent Assist systems.

Challenges in Development

Call Agent Assist systems face numerous hurdles during development:

Data and Privacy: Ethical collection and usage of customer data, coupled with adherence to stringent privacy regulations like GDPR and CCPA, pose significant challenges. One key challenge faced in this area is to clean the speech to text data to remove any confidential information before using that for fine tuning or RAG implementations.

Accuracy and Bias: Training AI models on diverse datasets to minimize bias and ensure accurate results across languages and regions is crucial. Capturing cultural nuances and communication styles adds complexity. There are lot of tactical challenges like tone detection, pitch to gender determination, speaker role determination etc. to overcome to achieve the desired level of accuracy.

Integration and Workflow: Seamless integration with existing call center infrastructure is vital for agent adoption and effectiveness. Overcoming data silos and outdated systems presents integration challenges. But even before that, proving reasonable correlation between agent assist scoring and the customer experience (call feedback survey/ payment rate/ transaction rate etc.) is the very first step to get the necessary buy in for the workflow integrations, which can sometimes be quite challenging.

Human Oversight and Explainability: Ensuring human control over AI decisions and transparency in recommendations is essential for building trust. Incorporating Explainable AI (XAI) techniques aids in understanding AI-driven conclusions. Effective implementation of RLHF (Reinforcement Learning for Human Feedback) can be useful to solve this problem. 

Scoring Universality and Trust

Scoring: Achieving universality in scoring is challenging due to language, cultural, and communication differences. Employing a multi-layered approach with language-specific models and cultural awareness can enhance accuracy.

Manual Scoring: While manual scoring can offer targeted feedback, it’s susceptible to subjectivity, bias, and time constraints. Combining manual scoring with AI-driven analysis provides a comprehensive evaluation framework.

Open-Source Datasets: Leveraging open-source datasets such as Switchboard, Ubuntu Dialogue Corpus, and CallMiner facilitates training. However, adaptation and pre-processing may be necessary to suit specific use cases.

Build vs. Buy

Build:

Pros: Offers customization, control over data and algorithms, and potential cost savings in the long run.

Cons: Involves high development costs, requires technical expertise, and entails ongoing maintenance and updates.

Buy:

Pros: Enables faster implementation, lower upfront costs, and access to pre-built features and expertise.

Cons: Provides limited customization, may result in vendor lock-in, and could entail a lack of control over data and algorithms.

Implementation Cost Estimation

Estimating the cost for in-house development depends on various factors, including the scope of the project, technical requirements, and expertise needed. A basic system with two data scientists and technology costs could range from $100,000 to $500,000 in the first year. However, actual costs may vary significantly based on specific project needs.

The global AI-powered call center market is expected to reach $3.9 billion by 2027, reflecting the growing demand for personalized customer service and agent productivity.

Conclusion

In conclusion, the development of Call Agent Assist systems is a journey through a complex landscape, rife with challenges and strategic considerations. From the crucial need to uphold data privacy standards to the intricate task of mitigating bias, each step in this process demands careful navigation. Moreover, the decision to build an in-house solution or procure an off-the-shelf product adds another layer of complexity, compelling organizations to balance customization against cost and time constraints.

However, amidst these challenges lies immense opportunity. Call Agent Assist systems, powered by Gen AI, are at the forefront of industry innovation. Beyond merely enhancing call center efficiency, these systems have the potential to revolutionize customer service by improving agent performance, elevating customer satisfaction, reducing call times, and ultimately enhancing the organization’s brand image. Moreover, their versatility extends beyond the confines of call centers, making them valuable assets that can be leveraged across various departments within an organization to address a multitude of queries and streamline internal processes like HR Shared Services. By proactively addressing these challenges and leveraging available resources effectively, businesses can harness the transformative power of AI to drive tangible improvements in customer experience and agent productivity.

This article is written by a member of the AIM Leaders Council. AIM Leaders Council is an invitation-only forum of senior executives in the Data Science and Analytics industry. To check if you are eligible for a membership, please fill the form here.