From CDR data to
margin intelligence.
In 5 To 7 Days.
One data source. Four AI models. A continuous intelligence layer that runs above your existing platform — delivering insights before your BSS knows there's anything to report.
What happens between "go" and first insight.
Margin Intelligence doesn't replace your existing platform. It connects to it, learns from it, and runs a layer above it. Here's exactly what that process looks like — and what you get at each stage.
CDR Feed Connected
Your CDR data stream is connected to the Margin Intelligence pipeline. Three connection methods are supported — direct MNO API (preferred, real-time), SFTP from your existing platform, or scheduled file upload. Any CDR format is accepted and mapped automatically during onboarding.
- Real-time API connection: CDRs enter the pipeline as they're generated by the MNO
- SFTP or file upload: processed on arrival, typically hourly or daily depending on your existing setup
- Format mapping handled by TelcoEdge — no changes required to your CDR output configuration
Rate card uploaded
For operators with a direct MNO wholesale relationship, your rate card is uploaded once. TelcoEdge maps it against your CDR plan codes automatically — no manual mapping required. When the MNO updates rates, the rate card is updated in the platform and the mapping reapplies immediately. MVNA-model operators skip this step — CDR Intelligence and Anomaly Detection run directly from CDR data.
7-day AI training run
The four AI models train on your last 7 days of CDR history. This establishes your baseline — what normal usage looks like for your subscriber base, what your typical cost accrual pattern is across a billing cycle, and what pool utilisation looks like across your wholesale allocation. You receive your first margin recovery estimate before the models go live — so the value is visible before the first billing cycle completes.
- Invoice Prediction: learns your CDR-to-cost conversion pattern across plan codes
- Anomaly Detection: establishes the normal baseline — deviations from it become alerts
- Pool Optimisation: profiles 7-day utilisation per pool allocation
- CDR Intelligence: learns subscriber-level usage patterns and activity baselines
All models live dashboard active
The Margin Intelligence dashboard becomes active. All four AI models are running continuously. Anomaly Detection fires on every incoming CDR batch. Invoice Prediction updates the daily accrual view. Pool Optimisation produces its first weekly recommendation. CDR Intelligence begins scoring every active subscriber for churn signals, upsell moments, and compliance windows. Your existing BSS continues operating exactly as it did before.
- Projected wholesale invoice available before invoice generation
- Real-time insights into cost, usage trends and margin performance
Continuous improvement
Every billing cycle adds to the models' training data. Invoice Prediction accuracy improves as it accumulates more completed cycle history. Anomaly Detection becomes more precise as the subscriber baseline deepens. The models are retrained monthly — the longer Margin Intelligence runs on your operation, the better it performs for your specific subscriber profile and MNO relationship.
- Invoice Prediction: target accuracy below 5% error rate — typically reached by billing cycle 3
- Anomaly Detection: false positive rate falls as the subscriber baseline matures over 30 days
- CDR Intelligence: churn prediction improves significantly after 90 days of labelled subscriber history
What you need to provide
CDR feed access
PI endpoint, SFTP credentials, or scheduled file upload. Any CDR format.
Rate card
for direct MNO operators only. Spreadsheet or PDF. Loaded once, mapped automatically.
Operator model confirmation
direct MNO or MVNA. Determines which modules activate.
What you see when the dashboard
goes live.
Four streams of intelligence, updated continuously. Each one addresses a gap that your existing platform doesn't surface.
<5%
Invoice prediction error
99%
Leakage detection rate
94%
Pool recommendation confidence
91%
Churn prediction accuracy
