Responsible AI delivers better outcomes
At TELUS Digital, we're big proponents of responsible AI.
In the past, we've outlined important responsible AI principles to follow that can deliver better AI outcomes not only for ourselves, our clients and our communities, but for all AI data service providers.
Since our inception nearly 20 years ago, we have been consistent in speaking to the importance of privacy, security and humanity-in-the-loop, but beyond words, we have applied these principles, embedding them into the decisions we make and the work we do in support of global and disruptive brands.
Helping to achieve a global first in privacy
With the amount of attention and investment garnered by generative AI (GenAI) today, achieving a 'first' in the field is a remarkable feat. But TELUS, our parent company and client, with support from TELUS Digital and Microsoft, managed to do exactly that.
In May 2024, TELUS, a leading telecommunications and technology company, made history with their GenAI customer support tool by becoming the first in the world to be internationally certified in Privacy by Design (ISO 31700-1).
The GenAI tool's evaluation was conducted by KPMG, an international auditing and advisory firm, which reported its alignment with international privacy criteria and the requirements laid out in the ISO 31700-1 Privacy by Design Standard. This achievement underscores our unwavering commitment to the highest standards of privacy and data protection, while continuously innovating to deliver a best-in-class customer experience.
Responsible AI highlights:
- TELUS GenAI was built prioritizing data privacy at every layer of development, adhering to ethical Privacy by Design and responsible AI principles.
- The tool has undergone rigorous risk mitigation testing to identify, evaluate and safeguard against vulnerabilities.
- TELUS also applies a collaborative and rigorous approach to testing its AI tools to reduce risks and improve accuracy, called Purple Teaming. This approach combines both adversarial testing, commonly known as red-teaming, with blue-teaming, that consists of innovative defensive techniques to effectively enable the company to identify weaknesses and confirm validity and robustness of an AI application.
- TELUS provides clear information on how the tool works, and how it doesn't work, while also following a data ethics framework and monitoring the system with human oversight.
Since roll-out began, the GenAI support tool has already answered more than 50,000 customer queries, helping 28% more customers find the information they are looking for versus conventional site search on their own. For added convenience, customers with account-specific questions can be directed to agents who are always available to help with more complex, personalized service requests. Customers have the ability to access fast and intuitive responses to their queries, providing them with a more convenient and seamless digital experience.
The TELUS GenAI support tool was provided by Fuel iX — an enterprise-grade AI engine designed by TELUS Digital to help companies deploy and manage customized GenAI solutions at scale faster — and integrates large language models from Microsoft Azure OpenAI Service. Fuel iX is one of three offerings from TELUS Digital that help brands build, fine-tune and deploy AI initiatives responsibly. Read on to learn more about these three offerings.
Fuel iX
Fuel iX, an enterprise-grade AI engine designed to help companies upgrade their generative AI pilots to production scale, deploy customized solutions faster and effectively manage the technology responsibly.
The AI engine enables existing enterprise applications, third-party and custom GenAI apps to access critical services like access control, moderation, observability and privacy through an integration server and GenAI gateway, simplifying management of the GenAI value chain from data and models, to management, applications and interfaces.
This functionality is further supported by TELUS Digital's design services, which provides custom applications and other digital solutions, as well as the GenAI Jumpstart accelerator program for rapid generative AI development and deployment.
Fuel iX has delivered impressive results:
- For consumer robotics company iRobot, Fuel iX powers an agent-assist bot that has been part of improving agents' confidence and reducing new-hire average call handle time (AHT) by more than 10% (and up to 25% in the first five weeks).
- TELUS generated $6.4 million (CAD) in cost avoidance and $1.8 million in revenue lift from more accurate billing for technician visits after implementing Fuel iX.
Fuel iX was created to provide enterprises with generative AI flexibility, control, productivity and trust. Our first customers were key divisions within TELUS and we were able to implement AI solutions ranging from IT service desk, field service and internal team member knowledge solutions to in-store retail associate copilots, customer service agent assistant and a public-facing consumer website — doing so safely, responsibly, and in record time. To date, we have over 30,000 users within TELUS, and these solutions have significantly increased employee productivity.GenAI Jumpstart
With a path-to-production focus, GenAI Jumpstart is an eight-week engagement designed for companies at an early stage of their AI journey, helping them to rapidly identify use cases, build powerful risk mitigation tools and deliver a functional GenAI-powered virtual assistant prototype.
A key feature of GenAI Jumpstart is the unique Dual-LLM Safety System. The system uses a large language model to supervise the results of a retrieval augmented generation (RAG) system. A RAG-based system runs on a company's private and secure database of controlled and secured information rather than the open internet to help ensure that responses generated by a virtual assistant only use approved information that conforms to regulatory frameworks. Unlike traditional chatbots that can struggle with maintaining up-to-date information or accessing domain-specific knowledge, this feature helps keep AI assistants focused to mitigate inaccuracies, hallucinations and jailbreaking — a form of hacking that aims to bypass or trick an AI model's guidelines and safeguards to misuse or release prohibited information.
See how GenAI Jumpstart is making a difference: TELUS Digital and WillowTree, a TELUS Digital Company, designed and developed a next-gen chatbot experience for a leading North American multinational bank in eight weeks that balanced its need to improve customer service and productivity with the industry's strict governance demands. The fully-automated Dual-LLM Safety System, which eliminates the need for data scientists to manually create and evaluate hundreds of prompts and responses, saves significant time. By leading research and testing with the bank's users, the team was also able to establish customized user experience guidelines for the chatbot. Following its successful delivery, the GenAI Jumpstart offering was codified to create a safe, secure, modular architecture that can be adapted to address use cases spanning all industries.
For many use cases and specific tasks, particularly in highly-regulated industries but frankly a necessity across all industries, organizations must strive to retain total control over the knowledge database that an LLM references. Our global team has the experience to take clients from whiteboard to working prototype over the course of an eight-week sprint, without sacrificing security and quality, and decreasing both time to market and critically, time to value.Fine-Tune Studio
Fine-Tune Studio is a multimodal and multilingual platform that enhances your fine-tuning journey with high-quality and diverse datasets powered by TELUS Digital's specialized global workforce to produce human-aligned AI outcomes. Honing, as delivered by Fine-Tune Studio, is essential in reducing bias in AI and in developing safe products.
Fine-Tune Studio creates high-quality datasets for LLM fine-tuning in more than 100 languages and can be deployed across domains where specialized responses are required, including science, technology, engineering and mathematics (STEM), healthcare, finance, law, coding and more.
Aligning with the paradigm shift to multimodal AI use cases, Fine-Tune Studio supports labeling tasks combining various data types like text, audio, image, and video for supervised fine-tuning (SFT), reinforcement learning through human feedback (RLHF), direct preference optimization (DPO) and red teaming.
It is more critical than ever that companies make responsible decisions to ensure the accuracy, diversity and integrity of the datasets used to train their models when scaling their GenAI use cases from concepts to pilots to production and beyond. With the launch of Fine-Tune Studio, we have taken another significant step forward in our ongoing efforts to develop proprietary technology that supports our ability to deliver expertly-vetted and validated data that improves our clients' model performance, adaptability, and safety.Innovate responsibly with a trusted AI partner
Looking to make history?
Partner with a recognized leader in AI and deliver better customer outcomes that maintain data privacy and security. TELUS Digital has over 19 years of experience supporting global, disruptive brands and has the in-house expertise to provide guidance around responsible AI.
