# Decoding the telecom data goldmine: CDRs, network telemetry, and usage signals
At 7:42 a.m., Maya unlocks her phone and checks the weather. That single tap generates a data session record. When she calls her sister at 8:15, her phone produces a call detail record. As she drives to work, her connection hops across four cell towers, each handoff logged in milliseconds. By the time Maya reaches her desk, she has left behind hundreds of data points, and she has not spoken to her operator once.
Every telecom operator sits on this exact stream, multiplied by millions of subscribers. This lesson traces Maya's day to show what gets captured, where it lives, and which business questions each signal answers.
Telecom data falls into three families. Learn these and most of the industry's analytics vocabulary clicks into place.
A CDR (Call Detail Record) is a structured log created every time a subscriber uses a billable service: a voice call, an SMS, or historically a fax. It is not a recording of the conversation. It is metadata: who, when, how long, and from where.
A simplified voice CDR looks like this:
calling_number: +1-555-0142
called_number: +1-555-0198
start_timestamp: 2026-03-14T08:15:03Z
duration_sec: 212
cell_id_start: 41072
cell_id_end: 41093
call_type: MO_VOICE # Mobile Originated
termination: NORMALCDRs were born for billing. If Maya's plan charges per minute, the duration_sec
Business questions CDRs answer:
termination: ABNORMAL) clustering in one region?For a plain-language reference on how these records are standardized, the ITU (International Telecommunication Union) publishes technical recommendations that operators build against.
Telemetry means automated measurement data streamed from network equipment: cell towers, routers, and core servers. Where a CDR marks a discrete event, telemetry is a continuous heartbeat of network health.
When Maya's phone handed off between towers on her drive, each handoff (the transfer of an active connection from one cell to the next) produced telemetry: signal strength, latency, packet loss, and the tower load at that moment.
Key telemetry signals:
Business questions telemetry answers:
Telemetry is high volume. A single tower can emit thousands of measurements per second, which is why operators increasingly process it as a real-time stream rather than storing every raw reading forever.
Most of Maya's activity is not calls. It is data: apps, video, maps, messaging. Each is captured in a data session record, sometimes called an xDR or an IPDR (IP Detail Record).
A data session record logs the volume of data moved, the duration, the network type (5G, 4G, WiFi offload), and often the category of traffic. Crucially, operators typically see the *type* of traffic (streaming video, web browsing, software update) even when the content itself is encrypted, because they infer it from patterns, endpoints, and volume.
Business questions usage signals answer:
Here is how the three data types stack up across Maya's morning.
| Time | Action | Data captured | Business use |
|------|--------|---------------|--------------|
| 7:42 | Checks weather app | Data session (small, 4G) | Usage profiling |
| 8:15 | Calls sister (3.5 min) | CDR + telemetry across 4 towers | Billing, coverage QA |
| 8:40 | Streams podcast in car | Data session (large, handoffs) | Capacity planning |
| 9:05 | Call drops near office | CDR (ABNORMAL) + low RSRP telemetry | Coverage gap alert |
| 12:30 | Video call at lunch | Data session + latency telemetry | Quality of experience |
Notice the pattern. No single record tells the full story. The dropped call at 9:05 is only *explained* when you join the CDR (an abnormal termination) with the telemetry (weak signal at that location). This joining is where telecom analytics earns its value.
Telecom data has three properties that make it a goldmine.
It is continuous. Unlike a retailer who sees you only at checkout, an operator sees Maya almost every waking minute her phone is on.
It is spatial. Every record carries location, at least at the cell level. This enables everything from traffic planning to advertising, and it is exactly why it is regulated.
It is predictive. Behavior changes often precede decisions. A subscriber who stops using data, calls a competitor's support line, and searches "cancel plan" is signaling churn before they act.
That power comes with heavy legal weight. In the EU, the GDPR (General Data Protection Regulation) treats location and communications metadata as personal data requiring a lawful basis to process. Regional rules such as the ePrivacy framework add telecom-specific constraints. In the United States, CPNI (Customer Proprietary Network Information) rules from the FCC restrict how operators use the data generated by a subscriber's use of their service. The practical rule of thumb: just because you *captured* a signal does not mean you may freely *use* it for any purpose.
🎬 [VIDEO: "How Mobile Networks Work" — youtube.com — A clear animated primer on cell towers, handoffs, and how your phone stays connected as you move]
Knowledge check
1. A telecom analyst describes a CDR as 'metadata, not content.' What does this distinction most accurately mean?
2. Although CDRs were originally created for billing, they now support many other functions. What does this best illustrate about telecom data?
3. A fraud team flags an account showing a sudden spike in international calls at 3 a.m. Which field in a CDR is MOST directly useful for building this kind of pattern-detection rule?
4. Select ALL correct answers about what a voice CDR typically records.
Select all the correct answers.
5. Select ALL correct answers describing business questions that CDRs can help answer.
Select all the correct answers.
Raw CDRs and telemetry are almost useless on their own. Value comes from three processing stages.
Billions of records get rolled up into subscriber-level and cell-level summaries. Instead of 4,000 individual data sessions, Maya becomes a profile: "45 GB per month, 70 percent video, peaks 8 to 10 p.m., strong home coverage, weak office coverage."
Raw records get joined with reference data. A cell_id becomes a neighborhood. A called_number prefix becomes "international, high fraud risk." A device identifier becomes "premium handset, upgrade eligible." Enrichment turns codes into meaning.
Enriched data feeds models that answer forward-looking questions:
A common beginner mistake is assuming operators run these models on raw records in real time. In practice most run on aggregated, anonymized, or pseudonymized datasets, both for performance and for compliance.
Two clocks run in telecom data.
Batch processing handles overnight jobs: billing runs, monthly usage summaries, churn scores refreshed daily. It tolerates delay.
Real-time processing handles what cannot wait: fraud blocking mid-call, congestion rerouting, and prepaid balance checks before a call connects. If Maya is on a prepaid plan and her balance hits zero, the system must decide within milliseconds whether to allow the next second of the call.
Modern operators increasingly push telemetry and usage signals through streaming platforms so that a coverage problem or a fraud pattern is visible in seconds, not tomorrow morning.