The observability company has rebuilt its data platform around the argument that enterprise AI fails on inputs rather than intelligence, and it is quoting internal test figures on token spend and time to knowledge that no external party has yet examined.
NETSCOUT SYSTEMS has expanded its data platform to sell packet-derived evidence as the operational context layer beneath enterprise AI, extending the deep packet inspection technology that built the company into the emerging market for agentic IT operations. The platform observes digital interactions, converts packets into compact contextualised evidence in real time, and curates that evidence at scale for observability, service assurance, cybersecurity and AI systems.
The company’s argument rests on a diagnosis of why enterprise AI projects stall in operations. Models keep improving, and they still produce unreliable operational decisions when the data reaching them has been sampled, aggregated or split across tools. Metrics, events, logs and traces, the MELT telemetry that most observability estates are built on, force an AI system to reconstruct what happened after the fact. That reconstruction costs inference cycles, compute and tokens, and it raises the probability of a wrong recommendation reaching an operator.
Sanjay Munshi, chief operating officer at NETSCOUT, put the position directly. “Unlocking the benefits of AI across the enterprise will not be achieved by adding another model. It will succeed through context engineering: giving AI the right operational context before reasoning begins,” he said.
NETSCOUT produces what it markets as Smart Data through two capabilities working together. Early semantic extraction derives operational meaning from packets at the point of observation, preserving evidence the company says disappears from conventional datasets once telemetry has been processed. Context optimisation at source delivers higher-density relevant context, so that AI systems consume less of their context window and compute budget getting to the same answer.
The numbers arrive without a baseline
Munshi attached two performance figures to the launch, both drawn from work inside the company. “Through our own internal testing we experienced more than a 25% reduction in AI token consumption compared with MELT-only data, and more than a 75% reduction in MTTK,” he said.
Neither figure comes with the detail an enterprise buyer would need to act on it. NETSCOUT has not disclosed the environment the tests ran in, the scale or composition of the workloads, the models used, the MELT configuration it measured against, or whether any external party reviewed the methodology. A 75% reduction in mean time to knowledge is a substantial operational claim, and its value depends entirely on what the starting position was. Buyers evaluating the platform should ask for the test conditions before treating either number as a planning input.
The analyst material NETSCOUT cites in support of its case came into the release from the company rather than from independent commissioning. Gartner has predicted that organisations prioritising semantics in AI-ready data will raise agentic AI accuracy by up to 80% and reduce costs by up to 60% by 2027.
IDC expects 80% of agentic AI use cases to require real-time, contextual and widely accessible data, and describes the objective as a trusted real-time environment where AI can reason, decide and act inside guardrails.
Governance is the part that matters as agents start acting
The commercial logic sharpens when AI moves from advising an operator to changing infrastructure without one. NETSCOUT is arguing that independently observed, explainable evidence is what makes that progression defensible, supplying the audit trail behind an automated action rather than an inference assembled from partial telemetry.
For CIOs in the region working under UAE AI governance guidance or Saudi Arabia’s SDAIA and NCA requirements, the ability to reconstruct why an autonomous system acted is closer to a compliance obligation than a technical preference.
Smart Data is embedded across NETSCOUT’s product line and can be pushed into customer data and AI workflows, which lets the company sell into estates that have already committed budget to other observability vendors. That positioning is deliberate.
As access to frontier models broadens and capabilities converge, NETSCOUT is wagering that the quality and token efficiency of the context supplied to those models becomes the durable point of difference, and that it owns a data source at the network layer that competitors reconstructing from logs cannot match.
Whether the wager holds depends on evidence the company has so far kept in-house.
