Why AI is creating a new divide in telecom

<h2><b><span class="h6">AI is reshaping telecom operations in real time, separating operators that adapt quickly from those still debating the business case.</span></b></h2>

<p>During my 26+ years in the telecommunications industry, I’ve watched as some operators adapt to change quickly, while others hold fast to familiar processes long after they’ve stopped working.</p> <p>Now, the gap between those two groups is growing wider every quarter, and AI is the reason.</p> <p>By adding AI-enabled operational intelligence to passive pipes, forward-thinking operators are tackling some of the industry’s toughest challenges. Yet too many operators continue to miss out on the business gains AI has to offer in trouble spots like customer churn, support and 5G monetization. Increasingly, the deciding factor<b> </b>in whether they<b> </b>make the move isn’t technology execution but leadership’s willingness to change.</p> <h3><span class="h4" style="font-weight: normal;">AI creates telecom networks that can think</span></h3> <p>For decades, telecom networks were, fundamentally, dumb pipes. They moved bits from A to B, and the operator sent you a bill. The intelligence lived at the edges, in routers, laptops and phones.</p> <p>AI flips this equation by moving the intelligence to the network. For operators, this is a huge leap toward autonomous networks, and according to industry results and our own conversations with operators, the early results are impressive. Operators report 30% to 40% fewer unplanned outages after deploying AI-powered predictive maintenance algorithms to detect towers at risk of failure. Many have seen mean time to resolve (MTTR) shrink from several hours to under 45 minutes.</p> <p>What’s making it possible for networks to operate more intelligently is agentic AI, which can flag problems, reason through them and take action. Some operators tell us their AI agents can monitor signal strength, track outages and convey real-time information to field teams before a human engineer has even opened a ticket. When customers report issues, the agents instantly cross-reference details such as live outage data and historical fault patterns to identify the source and recommend a resolution path.</p> <p>The KPI that tells the real story is network availability. Industry-wide, the standard for network availability sits at around 99.9%. Operators like Telefónica and SK Telecom, which have been running AI-driven network operations centers (NOCs) for years, regularly push that figure to 99.97% to 99.99%. That difference sounds minor unless you’re serving an enterprise customer whose SLA carries financial penalties for every minute of downtime.</p> <p>North American operators, including AT&amp;T and T-Mobile, are increasingly following the same path, with AI-driven NOCs becoming central to how they meet enterprise SLA commitments</p> <h3><span class="h4" style="font-weight: normal;">Telecom churn is a problem AI was born to solve</span></h3> <p>However, operational efficiency is only one side of the equation. Operators are also using AI to address one of their most persistent business challenges: customer churn. For many operators, churn is where AI starts delivering measurable business value. The industry’s 1.5% to 2.5% monthly churn rate may sound manageable, but over a year, it can mean replacing 20% to 30% of your subscriber base. In competitive markets, where acquiring a postpaid subscriber can cost $200 to $400, you’re burning capital just to keep pace.</p> <p>The conventional response has been to offer a discount to anyone who called to cancel. But AI enables operators to understand <i>why</i> customers are leaving in the first place, which allows them to get ahead of the problem.</p> <p>For example, applying natural language processing models to support calls, app reviews and chat transcripts can surface patterns no analyst would catch manually. Armed with the more granular data, machine learning-driven models can spot at-risk customers weeks before they leave. In some cases, we’ve seen operators identify at-risk customers with 75% to 85% accuracy up to 60 days before they cancel.</p> <h3><span class="h4" style="font-weight: normal;">Applying AI to telecom customer support challenges</span></h3> <p>Agentic AI can help telecoms finally tackle the biggest problems in customer support: high volume and slow resolution. The Tier 2 ticket backlog at major operators routinely contains thousands of unresolved issues.</p> <p>The operators moving fastest to address support challenges are deploying agentic AI directly into the stack. These systems can understand the full context of a customer’s billing history, cross-reference it against network events in their area, and resolve or escalate accordingly. For high-frequency, complex inquiries such as disputed charges and mid-cycle plan changes, agents can now manage end-to-end resolution without human intervention.</p> <p>By auto-triaging incoming tickets and redirecting them to the right resolution path, autonomous agents cut the back-and-forth that inflates turnaround time (TAT). Operators running these systems report 20% to 35% reductions in Tier 2 volumes within the first six months, with average TAT dropping 40% or more on targeted ticket types. This aligns with Gartner's 2025 data showing AI-driven support in telecom delivers an average 41% cost reduction on high-frequency inquiry types, including billing, outage updates and plan changes</p> <p>AI agents can also monitor data and pipelines around the clock—a big advantage in telecom, where persistent data quality issues require operators to staff dedicated analyst teams to watch for them. Now, autonomous agents can flag anomalies in real time and trigger remediation workflows before corrupted data propagates into downstream decisions like churn models and business reports.</p> <h3><span class="h4" style="font-weight: normal;">5G monetization: An AI problem in disguise</span></h3> <p>Despite its massive 5G infrastructure buildout, the telecom industry continues to face low-single-digit revenue growth, with operator revenues expected to grow <a href="https://www.delloro.com/news/worldwide-telecom-capex-to-decline-in-2026/" target="_blank">just 2%</a> annually. Enterprise 5G use cases continue to fall short of projections.</p> <p>Part of the problem is that 5G’s most valuable enterprise features—network slicing, ultra-low latency—require real-time orchestration that’s impossible to do manually. A human engineer can’t dynamically allocate dedicated slices to a hospital while simultaneously managing 80,000 fans streaming video at a stadium.</p> <p>But AI agents can. In our conversations with forward-thinking operators, they tell us they’re building networks of AI agents that each specialize in single domains, such as managing slice allocation or monitoring quality of service (QoS) thresholds. When a QoS agent flags a degrading enterprise circuit, the orchestration layer can reroute traffic, notify the account team, auto-generate a root cause summary and log a resolution recommendation, all without a human initiating the chain.</p> <p>One consumer division I worked with invested early in automated provisioning and activation. It’s the kind of unglamorous infrastructure work that rarely makes headlines, yet the division saw big gains. Within a relatively short period, provisioning times dropped to a fraction of the industry norm. Sales cycles shortened because delivery could now keep up with demand. Customer satisfaction scores climbed, and churn dropped. The investment compounded in ways the finance team hadn’t modelled for.</p> <p>The operators figuring out AI’s role in 5G monetization are positioning themselves to capture the B2B market 5G was always supposed to unlock. The ones that don’t could end up with very expensive infrastructure and no real way to differentiate on it.</p> <h3><span class="h4" style="font-weight: normal;">Leadership: the deciding factor for AI advances in telecom</span></h3> <p>The deciding factor in whether an operator implements operational intelligence is often leadership’s openness to rethink processes and operational models. I watched this play out firsthand within a client’s B2B division, where leadership continued to stand behind legacy systems and manual processes long after order failures had become a recurring problem. Provisioning was painfully slow, and maintenance costs were high.</p> <p>Meanwhile, the company’s consumer division had moved aggressively on automation and AI-driven support. The contrast between the two divisions’ outcomes was stark: B2B hemorrhaged costs and frustrated customers, while the consumer division grew. Same brand, same underlying infrastructure. The difference was leadership’s willingness to change.</p> <p>The risks of applying AI to these industry challenges are real, though. The clearest one is governance. When a single AI agent makes a decision, you can audit it. When a network of coordinating agents makes a collective decision, <a href="https://www.cognizant.com/us/en/insights/insights-blog/agentic-ai-risk-management" target="_blank">accountability gets murky fast</a>. Who owns the outcome? Whose logic do you interrogate?</p> <p>Operators getting this right are building structured environments where agents are onboarded with defined roles, governed by policy guardrails and monitored continuously for performance drift. The same discipline extends to data: GDPR fines in the EU telecom sector have already crossed €1.5 billion cumulatively, and the models powering churn prediction and personalization run on extraordinarily detailed behavioral data, raising real questions about data use and customer consent.</p> <h3><span class="h4" style="font-weight: normal;">Capturing the benefits of AI in telecom</span></h3> <p>The market lead among AI-adept operators is compounding, and compounding advantages are notoriously hard to reverse.</p> <p>As the global AI in telecom market heads toward a projected $38 billion by 2028, growing at about 40% annually, that capital is flowing to the operators investing in their businesses and away from the ones still debating whether AI’s return on investment is proven enough to justify it.</p> <p>The ones still debating won’t have much of a debate left to have by 2030.</p>