Network Intelligence: How Malaysian Telcos Are Deploying AI to Win the 5G Era

Malaysia's 5G deployment strategy has undergone a structural evolution that directly impacts how AI will be deployed across the nation's telecommunications infrastructure. The original Single Wholesale Network (SWN) model — where Digital Nasional Berhad (DNB) was the sole 5G network operator and all telcos accessed capacity as wholesale customers — was designed to accelerate coverage and avoid duplication of infrastructure investment. DNB achieved approximately 80% population coverage by 2024, with RM 4.5 billion invested in network infrastructure. However, the model faced persistent criticism from major telcos regarding pricing, quality of service guarantees, and the strategic limitation of being dependent on a single infrastructure provider for their most critical network layer. The transition to a dual-network model — where CelcomDigi (via its shareholder equity stake in DNB) and Maxis are authorised to build their own 5G networks alongside the DNB wholesale network — represents a fundamental shift in Malaysia's 5G architecture with direct AI implications. A dual-network environment means that AI-powered network management must operate across multiple infrastructure layers: the telco's own 5G network, the DNB wholesale network, and the legacy 4G/LTE network that continues to carry the majority of traffic. This multi-layer complexity is precisely the kind of operational challenge that makes AI not just beneficial but essential — no human operations team can optimise handover decisions, load balancing, and fault management across three concurrent network architectures in real time. CelcomDigi, with its combined subscriber base of approximately 20.7 million following the Celcom-Digi merger, has the scale to justify proprietary 5G infrastructure investment. The merger itself was driven in part by the recognition that AI-powered network operations, customer analytics, and enterprise solutions require the data volume and infrastructure investment that only a merged entity could sustain. CelcomDigi's stated strategy positions AI as a core pillar of its post-merger integration, with network intelligence, customer experience personalisation, and operational automation as the three primary AI investment domains. Maxis, with its focus on enterprise convergence and its MaxisONE platform, is approaching 5G-AI from the enterprise solutions angle — positioning its network as an AI-ready infrastructure platform for business customers rather than primarily as a consumer connectivity service. This strategic differentiation creates distinct AI development priorities: Maxis invests heavily in enterprise-facing AI capabilities (IoT orchestration, private 5G network management, edge computing platforms) while CelcomDigi's AI investment is more balanced across consumer and enterprise applications.

The transformation from reactive to predictive to autonomous network operations represents the most significant operational shift in telecommunications since the transition from circuit-switched to packet-switched networks. Traditional network operations centres (NOCs) operate in a reactive paradigm: alarms fire, engineers investigate, problems are diagnosed, and fixes are applied. In a 5G network serving millions of connections across hundreds of thousands of cells with sub-millisecond latency requirements, this reactive model is structurally inadequate. The volume of events, the speed at which degradation impacts customer experience, and the complexity of root cause analysis across virtualised network functions exceed human cognitive capacity. AI-powered network operations implement a three-stage maturity model. Stage 1 — Predictive — uses machine learning models trained on historical network performance data to predict equipment failures, capacity bottlenecks, and service degradation before they impact customers. Models analyse patterns across thousands of variables including traffic load, temperature, hardware age, software version, and weather conditions to generate maintenance and capacity planning recommendations with lead times measured in hours or days rather than the minutes available in reactive mode. Malaysian telcos implementing predictive maintenance report 25-35% reduction in unplanned network downtime. Stage 2 — Prescriptive — extends prediction to recommendation: when the system predicts a capacity bottleneck on a particular cell cluster during peak hours, it not only alerts the operations team but recommends specific actions — load rebalancing across adjacent cells, temporary capacity augmentation, or traffic steering to alternative network layers. The AI system evaluates multiple intervention options and ranks them by predicted effectiveness, implementation risk, and customer impact. Stage 3 — Autonomous — closes the loop: the AI system predicts issues, determines optimal interventions, and executes them without human approval for pre-authorised action categories. This is the operational model that CelcomDigi's Sophia platform is progressively building toward across its 300+ managed processes. Autonomous operations are essential for 5G network slicing — where the network dynamically allocates dedicated virtual network segments for specific enterprise customers or use cases with guaranteed performance parameters. Network slicing decisions must be made in milliseconds, at a frequency and complexity that no human operator can sustain.

In Malaysia's hypercompetitive telecommunications market, customer retention and lifetime value optimisation are existential priorities. The cost of acquiring a new mobile subscriber is 5-7x the cost of retaining an existing one, and in a market where number portability makes switching frictionless, churn prediction and prevention represent the highest-ROI AI application for Malaysian telcos. AI-powered churn prediction models deployed across global telecommunications operators achieve 95-96% prediction accuracy — identifying at-risk customers 30-60 days before they churn with sufficient confidence to trigger retention interventions that demonstrably reduce churn rates by 15-25%. The architecture of a modern telco customer intelligence platform operates across four layers. The Data Unification Layer aggregates customer data from network usage records (call detail records, data consumption patterns, location data), billing systems (payment history, plan changes, complaint records), digital interaction logs (app usage, self-service portal activity, chatbot conversations), and external enrichment sources (credit data, social media signals where permissible under PDPA). The Analytics Layer applies machine learning models for churn risk scoring, customer lifetime value prediction, next-best-action recommendation, and micro-segment identification. The Activation Layer translates analytics outputs into personalised retention offers, proactive service recovery actions, and targeted upsell recommendations — delivered through the channel (SMS, app notification, call centre) that the model predicts will be most effective for each individual customer. The Measurement Layer tracks the incremental impact of AI-driven interventions against control groups, continuously refining model accuracy and intervention effectiveness. CelcomDigi's implementation of this architecture leverages the merged entity's combined data assets — the richest customer dataset in Malaysian telecommunications — to train models that neither Celcom nor Digi could have built independently. The merger doubled the training data available for customer behaviour models, improved segment-level prediction accuracy, and enabled cross-network usage pattern analysis that reveals customer needs invisible in single-network data. Maxis has invested in real-time customer experience monitoring that uses AI to detect service quality degradation at the individual customer level and trigger proactive recovery before the customer contacts support — a capability that transforms the support interaction from reactive problem-solving to proactive relationship management. For enterprise customers, telco customer intelligence extends to IoT account management: AI systems that monitor the health and performance of enterprise IoT deployments, predict connectivity issues before they impact business operations, and proactively recommend network configuration changes that optimise performance for specific IoT use cases.

The convergence of 5G and AI creates an entirely new revenue category for Malaysian telcos: Network-as-a-Platform (NaaP), where the telecommunications network is not merely a connectivity service but a distributed computing platform capable of hosting AI applications at the network edge. This model transforms the telco from a utility (selling bandwidth by the gigabyte) to a platform (selling compute, intelligence, and application capabilities by the use case). The strategic significance cannot be overstated: traditional connectivity revenue is under secular compression from competition and commoditisation, while platform revenue offers margin expansion and customer lock-in that connectivity alone cannot provide. Edge computing is the technical enabler of the NaaP model. By deploying compute infrastructure at cell tower locations and aggregation points rather than centralised data centres, telcos can offer AI inference with latency under 10 milliseconds — a capability that unlocks applications impossible with cloud-based AI. Manufacturing quality inspection that analyses product images in real time on the production line, autonomous vehicle decision-making that cannot tolerate cloud round-trip latency, augmented reality applications that overlay AI-generated information on physical environments, and retail analytics that process in-store video feeds without transmitting sensitive footage to the cloud are all applications that require edge AI. The Malaysian telco best positioned for NaaP is determined by three factors: 5G coverage depth (not just breadth), edge compute infrastructure investment, and enterprise solution capability. CelcomDigi's scale provides the cell tower footprint for edge deployment, but the company must invest in compute hardware at tower locations and develop the orchestration platform to manage distributed AI workloads. Maxis's enterprise focus gives it the customer relationships and solution design capability, but it must build the infrastructure depth to deliver edge computing at scale. U Mobile, as the challenger, may find edge computing partnerships — leveraging tower company infrastructure rather than building its own — the most capital-efficient path to NaaP revenue. For enterprise customers evaluating private 5G and edge AI deployments, the decision framework involves four considerations: latency requirements (does the use case genuinely need sub-10ms response?), data sovereignty (does the use case involve data that cannot leave the premises?), bandwidth economics (is it cheaper to process at the edge than to transmit to cloud?), and reliability requirements (does the use case need to function during internet connectivity interruptions?). Applications that score high on three or more of these criteria are strong candidates for edge AI deployment.

The regulatory environment for AI in Malaysian telecommunications operates across three overlapping frameworks that telcos must navigate simultaneously: telecommunications-specific regulation administered by MCMC (Malaysian Communications and Multimedia Commission), data protection regulation under the PDPA (Personal Data Protection Act), and emerging AI-specific governance frameworks being developed by the National AI Office (NAIO) and referenced in the Responsible AI principles. The complexity arises because AI applications in telecommunications touch all three frameworks simultaneously — an AI system that analyses customer call patterns for churn prediction is a telecoms service (MCMC jurisdiction), processes personal data (PDPA jurisdiction), and constitutes automated decision-making about customer relationships (AI governance jurisdiction). MCMC's regulatory focus has expanded from traditional spectrum management and licensing to encompass the digital services and AI applications that increasingly define the telco value proposition. The commission's Mandatory Standards for Quality of Service require that AI-powered network management systems maintain service levels that meet or exceed the standards previously achieved through manual operations — meaning AI deployment must demonstrably improve or maintain service quality, not merely reduce operational cost. This creates a regulatory floor for AI implementation quality that effectively prohibits the kind of minimally viable AI deployments that might be acceptable in less regulated industries. The PDPA framework, amended in 2024, has direct implications for telco AI applications. Telecommunications companies hold some of the most sensitive personal data in any industry — location history, communication patterns, browsing behaviour, payment information — and the amended PDPA's expanded definition of sensitive data, mandatory 72-hour breach notification, and enhanced consent requirements mean that AI systems processing this data must implement privacy-by-design principles at the architectural level. Specific implications include: AI models must be designed with data minimisation principles (using the minimum data necessary for the intended purpose), customer consent frameworks must specify which AI applications their data may be used for (not merely generic consent to data processing), and model training pipelines must include anonymisation or pseudonymisation stages that prevent individual re-identification from model outputs. For enterprise telco customers, the regulatory implications extend to their own AI programmes. Enterprises using telco-provided data (network analytics, IoT data, location intelligence) for their own AI applications inherit regulatory obligations that originate in the telco relationship. The data processing agreements, consent frameworks, and security standards that govern enterprise access to telco data must be designed with AI use cases explicitly in scope — a requirement that many existing enterprise telco contracts do not currently satisfy.

TechShift's Telecommunications AI Transformation Roadmap is designed for three distinct client profiles within the telecoms ecosystem: network operators seeking to operationalise AI across their infrastructure and customer management systems, tower companies and infrastructure providers exploring AI-driven asset management and NaaP revenue models, and enterprise customers leveraging telco AI platforms for their own business applications. The roadmap follows a three-phase structure calibrated to the telecoms industry's regulatory requirements, network complexity, and customer sensitivity. Phase 1: Assessment and Strategy (Months 1-3). TechShift deploys its ARIA Assessment framework customised for telecommunications — evaluating AI readiness across network infrastructure, BSS/OSS systems, customer data platforms, cybersecurity posture, regulatory compliance, and talent capability. The assessment produces a Telecommunications AI Blueprint: a prioritised roadmap identifying the 3-5 highest-ROI AI use cases, the data and infrastructure prerequisites for each, the regulatory compliance requirements, and the organisational change management plan. For operators, the Blueprint typically prioritises network operations AI (highest operational impact) and customer intelligence (highest revenue impact) as the first two implementation domains. Phase 2: Deployment and Integration (Months 4-12). TechShift implements the Blueprint priorities using its 6-week sprint methodology — deploying each AI use case through a rapid pilot phase followed by production integration. For network operations AI, this typically involves deploying predictive maintenance models across the most failure-prone network segments, implementing AI-powered capacity planning that feeds into the capital expenditure process, and building the anomaly detection systems that form the foundation of autonomous network operations. For customer intelligence, the focus is typically churn prediction model deployment, next-best-offer personalisation engines, and proactive customer experience management systems. Each deployment includes MCMC QoS compliance validation and PDPA impact assessment as standard governance gates. Phase 3: Platform Evolution (Months 13-18). With foundational AI systems in production, Phase 3 focuses on three strategic objectives: first, integrating AI capabilities across previously siloed domains (connecting network intelligence with customer intelligence to create unified experience management); second, building NaaP capabilities by deploying edge computing infrastructure and AI application hosting platforms; and third, establishing the internal AI Centre of Excellence that will own ongoing AI development, model lifecycle management, and regulatory compliance. The TechShift engagement concludes with an ARIA re-assessment that documents capability uplift and a strategic brief identifying the next horizon of AI investment — typically including advanced applications like dynamic network slicing, autonomous operations at scale, and AI-powered enterprise service creation platforms.