Telecom operators are pushing hard toward autonomous network operations, and agentic AI is central to that push. But new primary research from STL Partners, supported by Volt Active Data and Ocient, finds that data, not model sophistication, is what’s holding deployments back. Network data is traditionally sampled or aggregated to manage backhaul and processing costs, and that practice strips out the spikes, retries, and correlated failures an agent needs to catch a genuine anomaly instead of noise.
This report breaks down the two data challenges standing between operators and real-time agentic decisioning: processing the sheer volume of network data at scale, and making trusted, standardised context accessible across domains. Drawn from interviews with 11 senior operator and vendor leaders, it shows why data warehouses and lakehouses solve enterprise analytics but were never built for sub-second operational decisions, and what a data product strategy looks like when it has to serve both people and AI agents from the same foundation.
At a high level, real-time agentic decision-making requires two things working together: live event data describing what’s happening on the network right now, and known context, such as historical baselines and neighbouring topology, that lets an agent judge whether an event is a genuine problem or business as usual. Live event data has to be thorough and timely; context has to be reliable, discoverable, and governed as a proper data product. Centralised platforms remain essential for governance and long-term AI training, but they introduce latency the moment a decision has to happen while the network state is still live, which is why a hybrid architecture, combining central platforms with distributed processing close to where events are generated, is emerging as the answer.
For network architects, data platform leads, and technical strategy teams evaluating what AI readiness actually requires, this report replaces guesswork with a concrete framework for matching use case requirements to the right architecture, from non-real-time capacity planning to sub-second dynamic spectrum and load balancing decisions. The shift is from treating every event as fit for a central warehouse to mapping which live data genuinely needs full thoroughness and building the distributed, data-product-governed foundation that supports it. Read on to see how operators are approaching both problems today.