Navigating the challenges and opportunities of full automation
Modern mobile networks require moving beyond reactive management toward automated self-management through proactive techniques powered by AI and machine learning. Industry frameworks like TM Forum's Autonomous Network Framework define five automation levels, with Levels 4–5 representing fully autonomous networks that anticipate operational needs, self-optimize, and self-configure.
Tupl's Network Advisor embodies this shift — an AI-powered engineering platform automating diagnostic processes that traditionally consumed most engineer time. But the road to zero-touch is far from smooth.
The Rocky Road Towards Zero-Touch
Each detected network issue triggers manual investigation involving gathering performance counters from multiple OSS tools, comparing performance trends, verifying configuration data, consulting alarm logs and ticket histories, and documenting findings. This process repeats thousands of times per week across national networks. The bottleneck isn't data scarcity — it's the limited human capacity to interpret it efficiently.
Four Key Challenges
Legacy Infrastructure: CSPs operate hybrid network environments where physical, virtualized, and cloud elements coexist. Legacy platforms lack open APIs or depend on manual configuration interfaces that resist orchestration.
Data Quality and Silos: Automation depends on rich, high-quality, and harmonized data, but CSPs face fragmentation with information in isolated silos using different schemas and update cadences.
Skills and Culture: Zero-touch networks require multi-disciplinary expertise combining traditional engineering with AI, data analytics, and DevOps. Organizations need cultural transformation and trust in automation outcomes.
Standardization and Interoperability: Many automation platforms remain vendor-specific and non-interoperable, complicating end-to-end automation despite progress from TM Forum and ETSI ZSM.
Solutions and Enablers
Traditional assurance performs sequential checks; Network Advisor achieves AI-driven parallelization executing all checks simultaneously while maintaining transparency through decision flows.
Five Key Enablers
AI-Driven Root Cause Analysis: Uses supervised learning, unsupervised learning, and reinforcement learning so ML models replicate engineering reasoning, ensuring both precision and interpretability.
Closed-Loop Automation: Connects observation, analysis, decision, and action in continuous feedback cycles, aligning with TM Forum levels of automation for closed-loop capabilities across multiple services and domains.
Unified Real-Time Data Fabric: Requires streaming data pipelines, data federation layers, and semantic data models enabling unified visibility and consistent AI inputs.
Explainable AI (XAI): Provides transparency and human-understandable reasoning to build engineer trust and facilitate model validation.
Intent-Based Orchestration: Adopts intent-based orchestration frameworks using TM Forum's Open APIs and ETSI ZSM reference architecture as common language.
Proof of Concept
Simulation results demonstrate the efficiency gains from automation:
| Task | Manual (avg. min) | Automated (avg. min) | Savings |
|---|---|---|---|
| KPI & counter retrieval | 15 | 0.5 | 97% |
| Configuration validation | 10 | 0.5 | 95% |
| Neighbor relation check | 8 | 0.5 | 94% |
| Trend & correlation analysis | 20 | 1 | 95% |
| Documentation/reporting | 10 | 0.5 | 95% |
| Total per resolution | 63 | 3 | >95% |
Real-World Validation: U.S. Tier-1 Operator
Network Advisor was integrated into an existing OSS environment, connecting to PM counters, FM alarms, CM data, and trouble-ticketing systems. It continuously ingests and correlates data, applying AI-driven diagnostic models and rule-based reasoning.
Performance outcomes:
- 90% reduction in manual workload
- 2.5× increase in engineer productivity
- Faster, more consistent incident responses
- Improved diagnostic accuracy and reproducibility
Automation case distribution:
- R1 – Auto-closed (51%): Automatically identified non-actionable incidents
- R2 – Closed-loop (5%): Autonomously executed corrective actions
- R3 – Engineer-attended (5%): Complex incidents requiring human expertise
- R4 – Unknown root cause (39%): Anomalies with undetermined causes for AI training
Conclusion
Network Advisor exemplifies how AI-driven automation transforms assurance from a reactive, manual discipline into a proactive, autonomous system capable of continuous learning. Engineers focus on strategic, higher-value activities rather than repetitive diagnostic tasks.
The path toward Level 5 autonomous networks is rocky, but operationally viable. Combining intent-based orchestration, explainable AI, closed-loop automation, and a unified data fabric, telcos can systematically evolve toward self-optimizing networks that anticipate failures before they impact customers.
About Tupl
Tupl is the proven AI automation company for telcos. Powered by TuplOS, our AI-native automation system, we bring network and customer operations into one governed workspace, so your expert knowledge stays with you and grows. Since 2014, we've combined startup agility with deep telco expertise to help operators modernize operations, unlock autonomous networks, and deliver superior customer experience.