The Pragmatist's Approach to AI in Telecoms
In this article
The potential benefits of AI extend to almost every element of a service provider's organization, from reducing operational complexity to generating cost savings to improving customer service and issue resolution. However, serious challenges persist across important topics such as data privacy, security, sustainability and stakeholder buy-in.
Given the many complexities and pace of change within the AI industry, is it any wonder that many service providers are unsure where to start with AI?
For this reason, WWT recently convened a panel of AI experts to share their practical and realistic advice for service providers who are beginning the journey to AI adoption. Here's what we learned.
In the ever-evolving landscape of AI in telecoms, service providers stand at a crossroads: Should they buy off-the-shelf AI tools or commit to building their own bespoke AI solutions? This decision carries significant implications for the near- and long-term success of a provider's AI adoption strategy and, consequently, the subsequent transformation of their organization.
WWT recommends taking a practical approach to AI solution assessment and delivery that prioritizes targeted AI solutions that can deliver fast outcomes while accounting for the viability, maturity and scalability of your long-term AI and data strategies.
The cost of training your own hyper-custom AI model from scratch could very well balloon to $100 million or more — a massive investment that doesn't even consider the chain reaction it might cause in regard to a service provider's cybersecurity requirements, cloud commitments or carbon footprint. Moreover, we estimate that the total cost of ownership for a mature AI model will require between 10 to 19 million hours of cloud usage for training alone.
On the other hand, service providers opting to consume an off-the-shelf grade AI solution will likely find some value in the information retrieval and rudimentary analysis capabilities of the solution while maintaining low costs, but they will sacrifice risk-limited security guardrails.
The allure of building a custom AI tool lies in the unparalleled control and security it provides. Service providers can tailor solutions to ultra-specific applications, paving the way for innovation and differentiation in a competitive market. Founding Director at Khalifa University 6G Research Center, Merouane Debbah, says:
However, bespoke development introduces challenges, notably the ongoing AI skills shortage and the imperative need for pristine data. As AI systems rely on data to learn and expand, ensuring data cleanliness is paramount. Given the massive data generation in the telecom industry, managing this complexity demands strategic effort.
According to Marc O'Regan, CTO EMEA at Dell Technologies, the decision should be based on the specific outcomes you want to achieve with AI. "If you're looking at doing something very specific, very telco orientated — something that's going to drive innovation, drive a differentiation within the market and get you ahead of your competition — that's something that you would probably want to do in-house.
"Whereas if it's something rudimentary, I think you're looking then to buy that capability in. Which isn't to diminish it — it will help you achieve what you want to do, quickly and efficiently, with a high level of impact and return on investment."
While customization holds allure, the buy option remains a pragmatic choice for certain scenarios. Service providers often need to leverage existing tools, recognizing that not every aspect of their AI strategy requires a ground-up build. The key is finding a balance between readily available tools and dedicated solutions tailored for service providers.
The path chosen ultimately hinges on the strategic applications and challenges specific to each provider's business needs. The vision for an AI-powered telecom network, with a focus on data sovereignty, may necessitate a blend of both buy and build approaches.
Ultimately, service providers will have to consider the strategic applications and challenges around their specific business needs in order to decide the best route to AI adoption.
Regardless of the chosen path, data control emerges as a critical concern. Implementing AI without proper safeguards can pose cybersecurity risks, making it imperative to control data flow rigorously. Transparency in methods and promoting AI literacy within the organization become essential.
Additionally, AI tools can be vulnerable to attacks, wherein manipulated data can be leveraged to trick the system into making incorrect decisions. Likewise, a lack of transparency surrounding the methods used can mean that it's challenging to determine how and why an AI system reaches a particular decision.
Before applying AI tools to your business strategy, encourage a general uptick in AI literacy within the business, particularly in core roles. Users should understand the inherent risks — particularly to proprietary data — and know how to avoid them. You must also establish the necessary boundaries to manage the flow of data to and from any AI-powered tools, with proper governance and controls in place, before enabling access.
As one speaker on our panel so aptly advised, we need to think of AI in the same way we think of electricity: "Powerful when applied correctly, but dangerous without the proper safeguards in place."
In the quest for AI adoption, service providers need to envision a roadmap that balances buy and build approaches, aligned with broader business goals. The focus should shift from mere technology discussions to strategic business considerations, addressing the why, what and how of AI integration.
Ray Mota, CEO at ACG Research, suggests reframing the conversation:
Stakeholder buy-in becomes a crucial aspect, necessitating tangible inputs, actions and results to demonstrate the potential business impact of AI tools.
As far as a year ago, McKinsey predicted that service provider success in AI adoption would influence which organizations remained competitive and which fell behind — a compelling encapsulation of the "why" behind exploring the risks and benefits of AI adoption. But when it comes to getting stuck on the path to AI success, nailing the "how" and "when" of AI adoption are of equal importance.
According to the leading AI experts on our panel, here are some tips on how to get started, get unstuck and chart a path to sustainable AI success:
1. Learn from other industries
AI has the potential to disrupt practically every industry, introducing new use cases and capabilities that haven't been possible before. Merouanne Debbah explains:
By looking to other organizations for inspiration, service providers can ensure they are keeping pace with innovation while navigating common issues, and perhaps even uncovering novel use cases of their own.
For example, service providers should consider how use cases from different sectors — like training data-safe AI models in finance or reducing medical mistakes in healthcare — might be applied to the telecoms industry.
2. Start small
There's a lot of potential for AI to transform the service provider industry, but you can't expect to introduce chatbots, network optimization, operational efficiencies and everything else all at once and find success.
Consider using pre-trained models as a starting point, so you can understand where the most impactful applications exist within your processes. This saves the potential financial loss from developing a tool that may not significantly address your unique challenges. A pre-trained model will also allow users to experiment and become more AI-confident and literate.
3. Define use cases
Carefully evaluate AI hype versus reality. It's easy to become swept up in the excitement of new releases, but ultimately AI will only revolutionize your business processes if implemented in a place where it can address an existing challenge.
Begin by identifying a specific challenge, define what a solution would accomplish and then experiment with using AI to solve it — much in the same way telcos did during the automation push. This approach is more manageable and allows you to be more cautious and deliberate with your AI strategy.
"Understanding where you are in the AI journey, assessing your level of maturity and agreeing on a plan — these are all vital," explained Daniel Vale, Senior Vice President at WWT. "If you don't understand this, how can you make money from it? Once we have these capabilities, we can start orchestrating the costs, revenues and business cases."
4. Secure stakeholder buy-in
Arguably one of the harder aspects of any effective transformation is securing stakeholder buy-in.
Historically, service providers have been burdened by the need to create value and tasked with justifying their decisions at every step of the transformation process. From the implementation of 5G to the dawn of AI, it's clear that accelerating business processes and attaining the necessary funding is dependent on bringing the C-suite and stakeholders along for the ride.
"The way service providers position the discussion needs to change," said Ray Mota, CEO at ACG Research. "It needs to change from a technology to a business discussion. We should be talking about the why, what and how.
"Once you address those challenges at the board level, you need to map them to economic value. So the discussion isn't about AI — the discussion is about whether you're going to generate business value."
Ensure you gather and provide some tangible actions and results, as well as an example of practical business applications, so you're able to demonstrate the potential business impact of your chosen AI tools.
Service providers are undoubtedly feeling the pressure to embrace AI technologies and ensure that they remain in step with the overall market. Yet it's arguably more dangerous to jump into the AI space without first defining a long-term strategy and becoming educated on the associated opportunities and risks.
During our panel with the AI experts, their advice was clear: Whatever your goal, it's important to focus on the business problem you intend to solve rather than deploying AI for its own sake. AI technology should be viewed in terms of the surrounding ecosystem you're deploying, including the hardware platform, security, data management and the relevance of the model.
One important takeaway is to not lose sight of the practical applications. AI is an exciting, evolving technology, but there's no benefit to remaining purely theoretical or jumping in feet-first without direction.