AI for Telecom Installation Classification & Risk Prediction

In large-scale telecom deployments, poor installation practices can lead to customer complaints, repeat visits, and loss of trust. One of the UK's largest telecom operators approached us with a critical objective: identify high-risk installations before failures occur.

The Challenge

The client faced:

• Over 100K installations per year, with thousands leading to issues
• No structured record of installation types (manual labels, inconsistent)
• Lack of visibility into which installation practices increased failure risk
• Rising costs from repeated interventions and lost customer satisfaction

Our Approach

We developed a two-stage AI system to tackle the problem:

• Installation Type Classification

  • Collected a dataset of historical installation images
  • Trained a deep learning model (ResNet-based CNN) to classify installation types from photos
  • Achieved over 92% accuracy across 8 predefined installation categories

• Failure Risk Prediction

  • Combined image-classified installation types with historical failure records
  • Trained a probabilistic model (XGBoost) to estimate the likelihood of a failure per installation type
  • Built an internal dashboard with real-time probability estimates and flagging rules

Results and Impact

The dual-model system delivered strong results within the first 6 weeks:

• 52% reduction in failure-related rework in pilot regions
• 88% precision in identifying risky installations before technician dispatch
• Improved training and standardization by understanding which practices correlated with failures
• Data-driven prioritization of quality inspections and audits

Key Learnings

• Computer vision AI unlocks valuable insights from operational images already being collected
• Risk modeling based on historical data provides early-warning systems
• Technicians responded positively to score-based feedback rather than binary judgment

Looking Forward

The client is scaling the system across new installation types and exploring integration into technician mobile apps for live risk feedback.

This project showcases the power of combining computer vision and structured analytics to reduce costs, improve service quality, and create a safer, more reliable customer experience in telecom.