How Telus has laid the data foundations for GenAI success

Generative AI is poised to revolutionize industries, but the journey from concept to measurable business value hinges on a critical factor: data readiness.

In the latest TM Forum webinar, How to revamp data strategies for generative AI success, hosts TM Forum, Amdocs and Canada-based communications service provider (CSP) Telus shed light on why accurate, well-governed and secure data is key to unlocking AI’s transformative potential.

“Accuracy isn’t just a technical requirement; it’s the cornerstone for building customer trust and achieving operational efficiency,” noted Gal Schreiber, Product Marketing Director, Data and AI-Led Solutions at Amdocs. Companies like Telus are already harnessing data-driven AI solutions to reduce operational costs by 50% and boost customer satisfaction, proving that success starts with the right data foundation.

But operators do not have to wait until they have a perfect data preparation strategy in place to start using GenAI, according to Hezi Zelevski, Head of GenAI Services at Amdocs.

While GenAI can seem daunting at first, a good approach is to start with specific use cases and specific challenges. At the same time, “the end vision should be that we can look at all the data sources that you have on your customers and actually create some kind of a view of a customer,” he added.

Building blocks

Mark Newman, Chief Analyst at TM Forum, said many CSPs are at the proof-of-concept stage with GenAI, and are now in the process of putting these PoCs into production. GenAI “is causing operators to think really hard and deeply about whether the data strategy they have in place is good enough,” he noted.

Canada-based Telus is one example of a CSP that is already actively deploying GenAI throughout its business in order to drive operational excellence and reduce costs.

Hasan Jafri, Vice President of Engineering, Business Solutions and Customer Revenue at Telus, said the CSP decided early on to build its own GenAI-based platform to address some of the inherent flaws of existing GenAI solutions on the market.

“We quickly realized … that GenAI products are insecure. We don’t want all our data exposed out there,” he said, pointing to issues with hallucinations, data integrity and bias. “So our approach was, let’s build a secure and trusted capability that we could use internally.”

What began as an internal project evolved into a public product that Telus named Fuel iX. “We were trailblazers. We didn’t have the ability to go leverage a tool or a product,” Hasan said.

He added: “Fuel iX is a great tool. Enhance that with AI tools from Amdocs and other product companies, because I think those products are very unique and specific to the system … then Fuel iX brings it together from a journey perspective, because internal journeys cross multiple domains.”

The platform was integrated with all of the major GenAI engines that were available at the time, “and we’re continuing to grow that model. We believe that by integrating GenAI, Fuel iX, can significantly transform the way we do business,” Hasan said.

Owing to this approach, GenAI is already making an impact at Telus, with use cases including coding copilots, diagram generators for architectural efficiency, and technical support bots.

“From the customer order process to billing, to payments … each individual system has an AI capability, but then you bring it together using Fuel iX. So that’s an accelerator that one can use,” he said.

For example, Hasan said coding copilots are helping Telus to increase code development by as much as 30%, while a bot created using Fuel iX helps customer service agents to understand and address customer bills.

“It’s reduced our average handling time or call volumes. It analyzes the billing terms, aggregate amounts, defects, discrepancies. It makes that information clearly available to our agents to assist the customers,” he said.

Data preparation and strategy for GenAI

Like Hezi, Hasan also made it clear that a fundamental requirement of any GenAI product is to “make sure your data is good. So you have to really focus on your own data engineering and data management program and you may want to invest in training for all your people,” he said.

Hasan highlighted a number of aspects here including the quality of the data architecture, governance and risk management.

“You need data sources that enable the best understanding of proper context, data discovery and understanding impact analysis, and you need to facilitate good cataloguing. And that’s all part of metadata management,” he said.

The second area is quality. “AI is only as good as the quality and reliability of the data used. So we implemented data quality checks, templates, during intake to address any issues and ensure data consistency, quality and currency. We’ve implemented a lot of templates to ensure that, and that’s underpinned by governance,” Hasan said.

Gal Schreiber, Product Marketing, Director, Data and AI-Led Solutions at Amdocs, agreed that data quality and governance are among the key requirements for any successful GenAI project. She warned that through 2025, at least 30% of GenAI projects will be abandoned after the PoC due to poor data quality and inadequate risk controls.

“To lead in deploying generative AI across the enterprise, we must commit to robust data engineering practices, and it’s crucial that we train AI agents effectively, ensuring their outputs are clear and understandable, and most importantly, we must guarantee that these outputs are used responsibly. We need to safeguard the integrity of our application and the user trust,” she said.

Gal said the data accuracy “sets the baseline for how reliable and valid the responses from GenAI application can be, and mistakes in data can lead to misleading AI outputs affecting business decision and customer trust.”

Other building blocks, she said, are enrichment, or enhancing GenAI’s understanding of the prompt intent to generate more creative and insightful responses; fairness, to mitigate bias; security measures, to protect against potential threats; and governance, with clear protocols to help maintain quality and compliance throughout the GenAI life cycle.

Model approach

In terms of Amdocs’ own approach to GenAI, Gal said its offering is “comprised of many parts, but the most important ones [are] adaptive operation and data integration, along with foundation model tuning. Foundation model tuning involves adding specific telecom data into our models, and this help produce results that meet the uniqueness of the industry.”

Here, Gal is referring to Amdocs amAIz, a telco generative AI framework. Hezi noted that many interactions today are triggered by the customer, but the ultimate aim is to be more proactive and provide what he sees as the “next generation customer experience.”

“The logic we took is, we want to have some kind of what we call intent routing. We want to be able to build different types of agents that specialize in network, in sales, in care, in marketing, but we want to build it in a way that they can interact between themselves [to make it] transparent for the customer,” he explained. “We’ve built what we call atomic skills, different types of skills within the network, within the GenAI platform that can actually deal with use cases needed.”

At the same time, Hezi made it clear that GenAI does not replace everything. “GenAI relies on the fact that we have data and … we can generate predictions and use AI or ML for that.”

Meanwhile, Ramya Patel, Head of Data and AI business unit, Americas at Amdocs, was also keen to emphasise that GenAI is “not something to be afraid about. This is an excellent learning opportunity.”

Hasan agreed that cultural change can be a big barrier to overcome initially. “The one thing I told all my staff when we started down this path, and you could see the fear in their eyes, was: GenAI or AI isn’t going to take your job, but the person that knows how to use it as part of their day-to-day work will,” he said.

Learn it, get comfortable with it, and apply it, he advised, “and you’ll actually have an advantage over everybody else. People have taken that to heart … and they’re truly making a difference.”

To watch the webinar on-demand click here.