AI can cut telco opex near term, but productivity gains will have to wait – Omdia

Ever since OpenAI's ChatGPT burst into public awareness in late 2022, communications service providers (CSPs) – particularly the larger ones – have shown greater urgency to embed artificial intelligence (AI) and machine learning (ML) into their organizations. They don't want to miss out (or seem to be missing out) on the AI/ML know-how and large language models (LLMs) that made ChatGPT, a chatbot capable of human-like conversation, such a resounding success. 

According to a recent report from Bank of America (BofA), "deep integration of AI'' could ratchet up a CSP's return of capital employed (ROCE) in Europe by two percentage points (from current levels of around 6%) through greater operational efficiencies, higher productivity and smarter allocation of capex. BofA talks tantalizingly of "game-changing returns."

Forward-thinking CSPs agree that AI will be a strong growth enabler. South Korea's SK Telecom, arguably one of the world's most AI-minded telcos, expects AI to impact both consumer and enterprise markets significantly. Around 9% of SKT's revenue currently already comes from AI-related services. The target is to lift that to 36% by 2028, led by data centers and semiconductors. 

There's also a high expectation that clever AI algorithms, fueled by ML, will enable CSPs to trim their workforces even further courtesy of greater network automation and installation of chatbots using LLMs. BT's soon-to-depart CEO Philip Jansen reckons the UK-based Group can eliminate 55,000 jobs by 2030, of which 10,000 will be down to AI. 

AI/GenAI reality check

Despite excited industry talk of productivity boosts, reduced energy consumption, predictive maintenance and quicker time-to-market, the nearer-term AI upside for CSPs will likely reside in greater cost efficiencies related to customer-facing interactions. Only once the data management foundations have been laid will AI drive more revenue-generating opportunities.

This is the overriding conclusion from Omdia, a Light Reading sister company, in its report "State of Play of AI in Telecoms" (subscription required). In a survey of CSPs conducted in February 2023, Omdia found that two of the most "mature" AI use cases (meaning more likely to be deployed than others) related to customer care: "chatbots and virtual assistants'' and "customer experience." These are the low-hanging AI fruit.

By contrast, the more network-oriented use cases that require much heavier lifting on data management – think "anomaly detection" and "process automation" – are some way off maturity. It is, as Omdia stresses, early days for telco AI. AI-driven change is more likely to be incremental than transformational in the short term.

"AI/ML is only useful in well-defined use cases with high quality input data," notes the report. "A user cannot simply point an algorithmic wand at a data swamp of network log data to magically make things work better. Instead, operators need to pose specific questions and use high quality, service-related data." No surprise then, says Omdia, that CSPs tend to focus their initial AI efforts on reducing customer support costs in CEM (customer experience management).

Moreover, following the launch of ChatGPT, an exponent of Generative AI (GenAI) – defined by Omdia as the creation of something new, such as images, text, and music – CSPs have been at pains to either parade GenAI tools or at least show intention of developing them. According to another Omdia survey in mid-2023, most CSPs (56%) were investigating GenAI while a fifth were in the testing stage. Over 20% of CSP respondents said they were already active GenAI users.

Aside from telecom-specific use cases, such as helping computer code generation and assisting network teams with root cause analysis, Omdia expects CSPs to use GenAI for general-purpose tasks across different business functions. Creating written and visual content for sales and marketing, for example, or summarizing written and verbal content. "GenAI," says Omdia, "looks to be primarily a tool for boosting efficiency more than revenue generation, at least in the near term."

Movers and shakers

Light Reading has been monitoring closely CSPs' moves into AI and GenAI in the wake of ChatGPT. While there is strong AI emphasis on CEM cost reduction, GenAI adoption is ramping up too (in line with Omdia findings). Vodafone is one of the most enthusiastic GenAI investigators with numerous proof‑of‑concept trials already underway. These range from external and internal chatbots to supply chain management and responding to bid submissions from vendors. It's running its LLMs on Microsoft Azure's OpenAI, integrated closely with its existing Google Cloud and Microsoft Azure infrastructure.

Among the AI/GenAI first movers in Asia are China Telecom, NTT Group and Korea Telecom (KT). China Telecom has set up a dedicated AI subsidiary, China Telecom AI Co., backed by 3 billion Chinese yuan (US$424 million). Its scope includes AI software development, hardware sales and solutions. NTT Group is building out a GenAI service for corporate clients, expected to launch March 2023. KT has launched an LLM that allows enterprises to create AI services using their own datasets.

In the US, AT&T has launched "Ask AT&T," which uses ChatGPT functionality but runs in an AT&T dedicated Azure tenant designed to prevent data leakage. Initial use cases, primarily aimed at making employees more effective and productive, include software development, document translation, network optimization, assisting customer care agents, providing answers to HR questions, and generating summaries of meetings with action items.

Perhaps the most ambitious AI/GenAI endeavor by CSPs to date is the "Global Telco AI Alliance", the result of an MoU signed by Deutsche Telekom, e& (formerly Etisalat), Singtel and SKT in July. Under the MoU, the four CSPs agreed to make "joint efforts" to accelerate the transformation of existing telco business using GenAI and "develop new growth drivers through new AI-powered business models." 

The following month SKT announced an additional $100m (€92m) investment in Anthropic, a US-based startup specializing in AI/ML. SKT and Anthropic intend to build an LLM tailor-made for telcos that can support various languages, including Arabic, English, German, Japanese, Korean, and Spanish. According to SKT, learnings from its Anthropic collaboration will feed directly into the Global Telco AI Alliance. 

Omdia, however, sounds a note of caution about the four-way collaboration. Although it is always useful to have CSPs working together on new technologies, the analyst firm expects it will "take some time" before anything tangible comes of it. "Getting so many people to agree on things such as data models will be challenging," says Omdia.