The Detail Gap Between Telco Data and Autonomous Decisions
Operators are not short of network data. They generate petabytes a day. The problem is what survives the pipeline: 1-in-100 event sampling and 5-minute averages are built to spot trends, and they strip out exactly the spikes, retries and correlated failures an agent needs to tell a genuine anomaly from noise.
STL Partners interviewed 11 Tier 1 and Tier 2 operators, global NEPs and BSS/OSS providers, and reviewed 21 live agentic AI deployments announced around MWC 2026. Thirteen run on data up to several hours old. Six run on seconds-to-minutes data. Two run on sub-second data, and both are customer-facing fraud detection. Live network operations deployments running on sub-second data: zero.
This infographic maps where that gap sits for two audiences at once. For operators, anomalies hide inside five-minute averages and batch reports land after the network has already moved on. For NEP and SI architects, sampled telemetry cannot feed a decision layer that must act in milliseconds, and without a complete trail from observation to decision there is no basis for trusting autonomous action. The shared root cause is the same: full-fidelity data exists, but the decision layer that can act on it in time does not.
It also sets out the architecture the research points to: one foundation, two layers. Ocient holds full-fidelity telemetry at petabyte scale with no sampling. Volt enforces policy in single-digit milliseconds on complete event streams and records every decision. One knows what is happening now, the other knows what it means. A European Tier 1 operator already runs Volt as the decision layer in its 5G mediation stack, on complete event streams, at national scale.