Empowering Telcos with GenAI: Upselling Voice Bot
Deployed an AI-powered voice agent for upselling, achieving 50% ROI improvement and 4X-5X scale vs current call centers with natural, human-like conversations and multilingual support.
Built on Amazon Bedrock with Retrieval-Augmented Generation
50%
ROI improvement vs call center baseline
4×–5×
scale compared to current call centers
Overview
About the Customer
In the highly competitive telecom industry, upselling is a crucial strategy for driving revenue growth and customer loyalty. Currently telcos leverage different channels like email, outbound calls, SMS etc. & strategies to drive upsell. For large telcos, outbound calling is one of the key strategies to drive upsell as it is used to generate a high ROI and conversions.
However, in recent times the scaling Human agents for making voice calls is becoming increasingly challenging due to following reasons:
• ROIs are declining due to increasing person-power cost
• It's difficult to scale call center due to increasing call center & infrastructure cost, and workforce attrition
For telecom companies seeking to scale efficiently while controlling costs, AI-powered voice agents offer a game-changing solution. The key advantages of AI voice agents are as follows:
• Scalability: can handle a higher volume of interactions. Businesses can scale up or scale down customer engagement quickly and cost-effectively.
• Personalization: can deliver personalized, real-time recommendations without sacrificing a seamless, human-like experience.
We delivered AI voice agents for a leading Telco in SEA with the goal of scaling up upsell while improving ROI as compared to the call centers.
The Challenge
Business Challenges
- Deploying outbound AI voice agents for upselling in telecom presents several key challenges: • Natural, Human-like Experience: Ensuring the bot delivers smooth, engaging conversations while avoiding 'hallucination' (inaccurate or irrelevant responses). • Support regional language: Ensure that the agents are multilingual and can converse with the customers in their preferred language. • Low Latency: Maintaining real-time interactions to prevent delays in communication. • Customizable Voice: Tailoring the bot's persona, tone, accent, and speech style to regional preferences for better engagement. • Speech Variability: Background noise, speech nuances such as accents can make it difficult for the bot to accurately understand and respond, leading to potential misunderstandings. Language diversity and multilingual conversations are also an incremental challenge. • Compliance: Ensuring the AI communicates valid, up-to-date, and legally compliant information, particularly with promotions or offers, to avoid misleading customers. • Conversation Complexity: Ability to handle complex conversations that may deviate from typical interactions that you might expect from a conversation. • Lengthy Inputs & Responses: Managing long-winded customer replies that can hinder the agent from extracting information efficiently, and conversely, handling of lengthy agent responses and how the agent reacts when the customer interrupts mid spiel. The challenge is to create an AI voice bot that balances accuracy, engagement, speed, compliance, and customization while handling unpredictable speech patterns.
The Solution
Solution Approach
Our Voice Agent Framework is designed to provide seamless, natural real-time conversations. It begins with audio input, which is transcribed into text using advanced speech recognition. This text is then processed by a Large Language Model (LLM) to generate a relevant and thoughtful response. To ensure accuracy and relevance, we integrated Retrieval-Augmented Generation (RAG), enabling the LLM to access external information as needed.
The response is then converted back into speech through a Text-to-Speech (TTS) engine, ensuring a smooth and engaging interaction. The framework is fully integrated with telephony services for call facilitation and supported by backend systems that manage call scheduling, interaction data, and system actions like placing orders or creating tickets. This solution combines advanced speech recognition, AI-driven responses, and high-quality voice synthesis to create conversations that are both natural and efficient.
Conceptual Framework for Upselling Voice Agent
Key Features: The key features which we developed to ensure the success of this project are as follows:
• Background Noise Filtering: Isolate out background noise from the communication with minimum latency to ensure clear communication and minimize misinterpretation of context.
• Interruption Handling: Manages interruptions smoothly, ensuring clear turn-taking and maintaining conversation clarity.
• Conversational Flow: Structured to guide customers step-by-step, from introduction to offer confirmation and subscription validation and minimize off-topic conversations.
• RAG (Retrieval-Augmented Generation): Accesses a dynamic knowledge base, ensuring contextually relevant and up-to-date responses.
• System Integration: Integrated with other systems like Customer Value Management (CVM) systems, the agent recommends tailored upselling packages to boost revenue.
• Customizable Voice & Personas: The AI voice agent can be tailored with distinct personalities—enthusiastic, persuasive, or empathetic—and customized accents, such as a natural local accent, to enhance engagement and align with brand identity.
• Responsible AI Guardrails implementation: Implements safeguards to prevent toxic or inappropriate language, ensures customer data privacy, and mitigates data leakage risks.
AWS Services Used
Built on AWS
Bedrock – Foundation model inference and LLM orchestration
AWS OpenSearch – Vector search and knowledge retrieval
AWS RDS – Relational database for structured data storage
S3 – Data storage for audio files and interaction logs
Fargate – Containerized application deployment for scalability
AWS WAF – Web application firewall for security
Eventbridge – Event-driven architecture for system integration
Outcome
Business Outcomes
- Great Conversation Quality: The voice agent ensures clear, natural communication with minimal latency, providing an experience that's smooth and engaging for customers. This high-quality interaction helps maintain a professional and human-like feel, making it easier for customers to engage.
- Faster experimentation and iterations: Businesses can run more experiments, gather insights from a larger volume of interactions, and refine upselling strategies faster, leading to continuous optimization and improved results.
- Increased Conversion: By enabling companies to handle more customer interactions simultaneously, voice bots break the limitations of human agents, driving higher engagement. This greater reach leads to increased upselling opportunities and higher conversion rates, ultimately boosting revenue for telecom companies.
- Actionable Customer Insights: Through conversation transcripts and sentiment analysis, AI voice bots provide valuable insights into customer behavior and responses. Businesses can better understand why customers reject upsell offers, identify common objections, and refine their sales strategies to improve future conversions.
About OneByZero
OneByZero is a frontier Systems Integrator specialising in AI Coworker design, build, and deployment for regulated enterprises across financial services, telecommunications, and retail. We operate across ASEAN, India, ANZ, and Japan, with a presence in the United States, combining domain expertise in regulated industries with deep AI engineering capability.
HQ Singapore · Tech HQ USA · Australia · India · Indonesia · Malaysia · Philippines · Thailand · Vietnam
