Case Study
The energy sector is undergoing a massive transformation, shifting from traditional grids to digital-first service models. This change has redefined the customer relationship, where today’s energy consumer demands instant resolution for everything from reporting outages to managing billing.
For a leading multinational energy player, meeting these expectations meant modernizing their Contact Centre as-a-Service with conversational AI. The goal was ambitious: automate complex interactions to provide faster support. However, in the high-stakes utility industry, an AI misstep such as misclassification, hallucinations, bias, or privacy breaches amounts to a regulatory risk and a breach of trust. The company had to balance scaling an AI model that was intelligent enough to understand intent yet secure enough to protect sensitive user data. They needed a partner to bridge the gap between innovation and integrity. They turned to Infosys to implement a rigorous AI Assurance framework that ensured Quality Engineering (QE) rigor into every stage of the AI lifecycle.
98.9%
Accuracy in consolidating sensitive information
95%
Accuracy of responses across 98% of the knowledge base
200+
Edge-case security vulnerabilities identified and mitigated
95%
Accuracy in recognizing customer intent
Key Challenges
- The AI model struggled to accurately recognize customer intent, leading to misclassified queries and frustrated users
- The system was susceptible to 'jailbreaks' and adversarial attacks, where bad actors could manipulate AI into bypassing safety filters
- The AI system exhibited inconsistent behavior, hallucinations, fairness issues and ethical compliance risks due to unregulated LLM responses
- Manual testing was too slow to keep up with the volume of interactions, stalling deployment
Ready to experience?
Talk To ExpertsInfosys Approach
- Adversarial stress testing: Executed 800+ test runs across four iterative cycles, including specialized 'Red Team' probes to test the bot’s resilience against malicious inputs
- Privacy guardrails: Implemented rigorous privacy checks, achieving 98.9% redaction accuracy, ensuring zero critical leakages of sensitive customer data
- Automated simulation: Moved away from manual validation to a reusable automation framework, automating 90% of User Acceptance Testing (UAT) and customer journeys
- Intent optimization: Tuned the model to recognize customer nuances, reducing misclassification errors significantly
The Solution
Infosys transformed the client’s approach by moving from simple testing to comprehensive QE for the AI lifecycle.
Leveraging Infosys AI Assurance Platform, the team implemented AI Assurance framework. This framework ensured that client’s conversational AI from intent recognition to LLM-based knowledge retrieval, aligned with business objectives, sector regulations and ethical expectations
- Agent Benchmarking (Performance): We evaluated the AI system against industry standards, ensuring high-confidence responses
- Business Assurance (Alignment): We aligned AI’s behaviour with enterprise goals, ensuring it could handle real-world customer journeys accurately
- Red Teaming (Security): We played the role of the adversary, launching over 200 simulation attacks (from slang and typos to code-mixing and prompt injections) to find and fix vulnerabilities before they went live
- Responsible AI (Ethics): We embedded guardrails for fairness, transparency, and privacy, ensuring the AI system adhered to ethical standards and data protection laws
60%
of Tier-1 customer queries completely automated
30%
reduction in misclassification errors
40%
reduction in manual testing efforts
70%
faster onboarding for new AI models via the reusable framework
Benefits
Infosys secured the future of customer engagement for the energy sector, proving that in the world of AI, QE is the ultimate business
The client significantly elevated customer experience by reducing errors and automating routine queries, providing instant, accurate support during critical moments
The 'Red Teaming' approach ensured the client met stringent regulatory requirements, protecting the brand from reputational damage associated with AI bias or data breaches
The client now possesses a modular, reusable assurance framework. This 'blueprint for trust' allows them to rapidly roll out new AI features across different business units without reinventing the compliance wheel.
Request for services
Find out more about how we can help your organization navigate its next. Let us know your areas of interest so that we can serve you better.
