At the beginning of May, TELUS Digital officially rolled out its Fuel iX Agent Trainer tool across its 64 global delivery centers and digital studios. NelsonHall spoke with the Fuel iX team to learn about the latest deployments of the enterprise GenAI engine, the target business cases, and the next apps on the development roadmap.

AI platform for connecting infrastructure with LLM models

Fuel iX Core is TELUS Digital’s multimodal centralized management platform for GenAI applications. It has a library of LLM models from OpenAI, Google, Anthropic, Mistral, Meta, and Microsoft, and enables the creation of turnkey or custom GenAI applications. Fuel iX is hosted on Google Cloud.

TELUS Digital has created three app families so far:

• Fuel iX Copilots, general-purpose and no-code customizable copilots for enterprise employees
• Fuel iX Fortify, automated red-teaming
• Fuel iX for Customer Experience, currently including two generally available offerings: Agent Trainer and Agent Copilot. Another one, Customer Assist, for customer self-service support, is in production at scale for several TELUS products and will be made available more broadly later in 2025. 

Fuel iX is offered as a licensable platform for integration with third-party apps (e.g., turbopuffer), and includes RAGaaS, guardrails as a service, and orchestration services. The core layer controls access to the different RAG pipelines and LLMs, while the Fuel iX team creates application integration and additional functionalities with automated failover for high availability and easy switch between models, RAG, and guardrails.

With development beginning in 2019, the current incarnation of Fuel iX has five products live with TELUS Digital, its parent company TELUS Corporation, and its other subsidiaries. Using Fuel iX, TELUS and TELUS Digital employees have built 18k user-customized copilots, in addition to supporting 30 GenAI custom applications already deployed in-house and another ~30 in development.

AI trainer leveling up trainees

Fuel iX Agent Trainer creates AI-enabled simulations for CX agent training, roleplay, and learning in real time. Agent Trainer supports voice and chat simulations, configurable support scenarios, customizable skills assessment, automated performance evaluation feedback, and machine translation for multilingual programs. It automatically tracks and analyzes every training session with analytics to guide targeted coaching and the creation of personalized development plans. TELUS Digital is also using Agent Trainer’s conversational data and insights from individual interactions and processes to design optimized operational models and increase the speed to proficiency with performance quality enhancement to reach top-level CSAT for new hires.

An example deployment is with an online marketplace in the hospitality industry to transform the support training process. Agent Trainer was deployed for 242 agents and resulted in 25% to 50% faster speed to competency (the difference coming from queue type and language). Other implementations are with TELUS Digital’s Fortune 500 clients in retail, telecommunications, and financial services.

Accelerating productivity and quality in CX services with GenAI

The CX Services market is well into the large-scale deployment and ROI assessment stage of GenAI tools, but several of the initial challenges around scalability, reliability, and cost uncertainty remain. Enterprise clients are set on preventing LLM vendor lock-in, maintaining control and trust (e.g., avoiding probabilistic and irresponsible AI), while opening GenAI capabilities to frontline employees.

Fuel iX Agent Copilot operates in CX environments for knowledge, translation, and assistance, breaking down and targeting the different segments of customer interactions. Agent Copilot augments hard skills, actions, and processes; for example, via automated interaction summarization and more efficient knowledge discovery, retrieval, and recommendations.

An example deployment is for a multinational online marketplace, which struggled with an inefficient support system, with agents facing extended handle times and inconsistent responses due to knowledge base challenges. The Fuel iX Agent Copilot gave agents organized information, automated policy summarization, and suggested responses, resulting in decreased time for research by 42%, reduced AHT by 13%, and shortened time to resolution by 27%.

In another example, for a gaming studio’s player support program, the Fuel iX implementation unified fragmented knowledge across multiple titles and manual processes with automated ticket categorization, offering game-specific responses, and simplifying purchase verifications. TELUS Digital achieved a 30% decrease in time to resolution, a 49% decrease in AHT, and 11% higher accuracy in tagging.

GenAI customer bots: from deflection to personalization

Traditional Conversational AI allowed a number of support scenarios to be deflected and self-serviced but lacked personalization and contextual awareness. The new generation of GenAI-powered bots, such as Fuel iX Customer Assistants, are capable of listening, conversing clearly, understanding, researching, and resolving customer issues, albeit within a narrow field and specific processes, entering more closely the realm of agentic AI. TELUS Digital already has several deployments of Customer Assistants where the bots are knowledgeable and built with inference first. They lead 24x7 contextual and asynchronous customer conversations, document outcomes, reach into transactional systems, translate content, and even web scrape external resources to provide resolutions.

One deployment is with Koodo Mobile, a Canadian challenger brand, for which TELUS Digital implemented a hybrid GenAI and NLU in the existing virtual assistant to handle ~30k monthly interactions while reducing call volumes in key domains by 50%. The containment within the bot increased by 15%. At the next stage, the virtual assistant should handle ~150k interactions per month with an expected value add for the client of up to CAD 1m.

Scaling threat detection with Fuel iX Fortify

Another area where Fuel iX can assist humans is in the safety and security of customer-facing virtual assistants. The existing vulnerability detection tools are not designed to comprehensively evaluate bot responses to customers, while bad actors adopt GenAI to generate attacks against conversational bots at scale. The current manual process through red teaming cannot handle the volume and diversity of answers and relies on intervention techniques such as guardrails while overlooking preventive measures.

To address these challenges of achieving safe and ethical AI, TELUS Digital developed Fuel iX Fortify. Fuel iX Fortify is an automated red teaming tool for proactively detecting vulnerabilities in GenAI bots. The LLM-powered application simulates real-world adversarial attacks on GenAI-enabled chat assistants and copilots. It automates testing simulations on target applications by designing and launching a series of novel attacks, assessing their effectiveness within a judgment framework, and summarizing results. Non-technical users can fine-tune malicious attack settings and codes of conduct, and design scenarios for GenAI bot testing. Fortify covers 139 attack types across 13 risk categories, such as general and hate speech and harassment, government impersonation and subversion, misinformation, crime, privacy exploitation, and self-destructive activities.

Natural progression towards multimodal, multi-model, safe AI

The natural progression of Fuel iX’s adoption at TELUS Corporation and its subsidiaries has been executives witnessing the GenAI deployments in the CX services practice and looking to replicate it in other general-purpose horizontal areas or functional specialties such as technical account management ops, QE, enterprise client onboarding, sales planning, and marketing messaging. The direct benefits already experienced by TELUS have been raising the performance floor with the Fuel iX Copilots. The organization is wary of shadow AI and sees Fuel iX as the democratized platform for employees to easily self-service their GenAI needs.

The Fuel iX unified platform architecture enables further integration of Agent Copilot and Trainer to feed into Customer Assistants for improved self-service. One expected outcome of this agent augmentation tooling is reducing the requirements for hard skills by agents, shifting the focus to soft skills, and helping with more nuanced problems. TELUS Digital is already incorporating these AI accelerators into outsourcing client bids and talking to in-house operations as a licensing play and potential managed services entry. For its Customer Assistant, the company is more cautious, offering it to select enterprise clients first.

Another example of this synergy is Fortify, which benefits from the classifications and knowledge generated by Customer Assistants. With Fortify, the Fuel iX team is now targeting organizations without the means for sophisticated ongoing red teaming to lower the costs of ensuring responsible AI use with a variety and density of testing.