Telcos eager to leverage AI, but implementation challenges abound
This article was published in SDX Central on April 26, 2024. You can read the original article here
Telecommunications companies still struggle with numerous challenges — network optimization, latency, outages, customer service, data security and privacy just to name a few.
As with nearly all other industries, AI can help — so long as communication service providers (CSPs) can get a handle on the technology and its multitude of use cases.
“The rapid cadence of this technology’s evolution is something we’ve never seen before,” Anthony Goonetilleke, group president of technology and head of strategy at Amdocs, told SDxCentral. “This requires a paradigm shift in terms of adoption, experimentation, innovation and productization — an approach that’s separate and distinct from traditional rollout models that were suitable for prior technologies.”
Excitement is there, but so are big challenges
Recent Amdocs-commissioned research found that, while 90% of CSPs know that generative AI will have a profound impact on business goals, deployment is still in its early stages. Just 22% of CSP have implemented gen AI tools, 32% are running proof of concepts (PoCs) and 29% plan to explore it in the next 12 months.
Most providers consider data quality, regulatory compliance and large language model (LLM) training as the biggest hurdles.
“Data and privacy are two of the most critical considerations for service providers around the world; this is why the concept of trusted AI is a mandatory component to any solution rolled out in the telecommunications industry,” said Goonetilleke.
The largest global brands in the industry are responsible for managing the world’s most critical infrastructure, which requires careful consideration when rolling out generative AI. Also, the industry sits on a significant amount of data — and normalizing and ingesting that is a challenge in and of itself.
“Any successful generative AI deployment is dependent on the integrity and quality of the underlying data, which makes data a critical component of any long-term generative AI strategy,” said Goonetilleke.
Telcos also struggle when it comes to connecting the dots between “fragmented and siloed systems,” said Adil Belihomji, CTO at mobile networking company OXIO.
“Telcos today have built AI use cases on top of their already complex and layered systems and platforms,” he said. “As a result, they have multiple AI brains telling them how things should perform, instead of one connected experience.”
AI for network planning, operations and optimization
Top use cases for generative AI in telco include network planning, operations and optimization and fault resolution, says Richard Brandon, VP of Strategy at RtBrick, which provides multi-service edge routing software.
Network operators rely on historical data that can be difficult to interpret. AI can help them plan better by bringing “a new level of responsiveness” and correlating more factors, said Brandon. For instance, they can upgrade capacity in areas with increasing demand.
AI can also take advantage of real-time data “on a highly granular level” to help predict user behavior, he said. Today’s networks consist of thousands of nodes, various equipment types and millions of subscribers, so doing this task manually can be labor intensive and inefficient.
When it comes to fault resolution, meanwhile, AI tools can help detect failures that aren’t initially obvious. They can then identify complex root-causes to help avoid subsequent failures, and someday even predict and resolve faults before they happen.
Intelligent applications of generative AI can “propel telecom companies towards more proactive, rather than reactive, network management strategies,” said Raj Shah, managing partner and North America lead for telecom, media and technology at digital consultancy Publicis Sapient.
Scenario planning, improving customer experience
AI provides other unique opportunities such as digital twins, Belihomji pointed out. When applied for network equipment, digital twins allow for test cases that allow operators to better understand equipment behavior under specific conditions such as surge, load or failover.
Operators can run scenario planning in controlled environments, rather than risking out in live environments. The ability to multi-variant test and utilize a champion/challenger model — comparing multiple competing strategies — will provide “endless experimentation capabilities” when it comes to configurations, implementations and rollouts,“ said Belihomji.
”Meanwhile, co-pilot AI can give useful guidance on designs, scripts and configuration tools, and predictive AI can provide preventative care and take actions around high availability. Machines will begin to handle problems that “they’re better suited to solve,” said Belihomji.
This value will then pass through to customers, who will experience far less impact from deployments and upgrades. While it’s not emphasized as much, customer experience is a critical component of telecom revenue, Belihomji noted.
“Digitally savvy customers expect a personalized, 360-degree experience and AI can help us understand and anticipate their behavior and needs,” he said.
Goonetilleke agreed, saying the technology can help service providers generate more revenue by creating personalized and relevant offers, recommendations and promotions. It can also be used to help predict and prevent churn, handle billing-related inquiries and to recommend alternative products or add-ons. This, in turn, can call deflection rates and halve average handling times for assisted calls.
AI for back office and operations, too
But telcos should also be looking at opportunities in their back office and operations, Shah emphasized. “Some of these spaces actually present less risk for CIOs to start using generative AI,” he said.
For instance, AI can help optimize revenue management by analyzing billing systems, usage data and customer profiles to identify discrepancies that may be the result of leakage or billing errors.
In fraud detection, “dynamic learning algorithms” can adapt to new, evolving tactics and simulate fraudulent behaviors and predict their outcomes. For example, it can detect unusual patterns of behavior in real-time, such as the cloning of SIM cards or unusual spikes in usage that can indicate fraud.
“By recognizing these patterns early, telecom companies can intervene promptly, thus minimizing financial losses and protecting their customers from fraud-related disruptions,” said Shah.
Further, AI can help facilitate cross-system integration between operation support systems (OSS) and business support systems (business support system (BSS)), he said. This can provide a unified view of customer and network operations and help ensure that discrepancies between what is recorded in operations and what is billed are quickly identified and addressed.
At the same time, AI provides an opportunity for migrating OSS, BSS and other legacy systems into newer architectures and technologies. Code can often be written in older languages such as COBOL and often there isn’t supportive documentation, Shah pointed out. AI can analyze code for developers and help move to new systems, reducing maintenance costs and making way for newer innovations and capabilities.
Clearly, the use cases are extensive — and some haven’t even been identified yet.“As Telcos examine the impact of Generative AI on their businesses,” said Shah, “most are still just scratching the surface.”
Safeguarding sensitive data is paramount
While generative AI can no doubt bring “transformative change” to the communications industry, operators shouldn’t be hasty, said Goonetilleke. It requires a tempered, tactical strategy.
Notably, organizations must safeguard sensitive data and ensure the integrity and trustworthiness of AI algorithms. They should be implementing robust encryption, adopting observable AI techniques and adhering to stringent regulatory frameworks.
Fostering transparency in AI models and promoting ethical practices in data handling is also critical for building trust among users and stakeholders. This will become increasingly important as models expand with “billions and billions of additional data points,” said Goonetilleke.
“Coupled with paying close attention to emergent behavior and hallucinations, unleashing the power of generative AI requires an innovative and carefully crafted approach,” he asserted.