What edge to enterprise means and why it matters
Every industrial operation generates continuous streams of process data: temperatures, pressures, flow rates, vibration signatures, motor currents, product counts, batch parameters. For most of the history of industrial automation, this data lived inside the control system that generated it, accessible only to operators in the control room using SCADA screens. Enterprise systems, analytical platforms, and business decision-makers had no reliable access to real-time process data. The automation pyramid kept OT and IT in separate vertical silos by design.
Edge to enterprise is the architectural pattern that breaks down those silos without breaking the security and reliability properties that protect production systems. The pattern connects every physical asset at the field level to every analytical and enterprise application at the top level, through a structured, secured, quality-validated pipeline that preserves the meaning of the data at every step. When the architecture works correctly, an enterprise AI model can consume process data from 50 different PLC types across 12 sites in three countries through a single, consistent, ISA-95-contextualized data feed. When it does not work correctly, those same assets are data islands whose measurements never reach the systems that could act on them.
As of 2026, analysts and platform vendors consistently identify edge-to-enterprise connectivity as the primary gating constraint for industrial AI and IIoT value realization. Ericsson's January 2026 analysis of manufacturing AI programs finds that "most manufacturers have spent heavily on analytics platforms, AI development, and data science talent, yet many of these initiatives fall short, not due to immature technology, but because high-frequency OT data from critical equipment is siloed in legacy wired networks and fragmented device ecosystems that weren't designed for continuous, real-time industrial workloads." The edge-to-enterprise pipeline is the infrastructure that determines whether AI investments deliver returns or remain disconnected from the process data they require.
The 80% bandwidth reduction principle: Edge computing allows data deduplication and aggregation at the source. By transmitting only report-by-exception events or compressed summaries rather than continuous raw telemetry, industrial facilities typically achieve an 80% reduction in data backhaul costs. A 1,000-tag process unit generating 10 data points per second per tag would produce 10 million data points per minute if streamed continuously to the cloud. Report-by-exception transmission, where only changed values are published, reduces this to the subset of tags that are actually changing at any given moment, which is typically 5-20% of the total tag population at any time. The edge layer is what implements this filtering. Without it, cloud pipelines are consumed by noise from steady-state sensors that add zero analytical value.
The automation pyramid vs the edge-to-enterprise architecture
The traditional automation pyramid (the Purdue Model hierarchy from the 1990s) was not designed for data flow. It was designed for control isolation: each level of the hierarchy isolates the level below it from the network above. This was excellent for control security in an era before cyberattacks were a primary threat, but it was deeply hostile to data access. Moving data from Level 1 (sensors/PLCs) to Level 4 (ERP) required it to pass through every intermediate layer sequentially, with translation at each boundary. The result was slow, expensive, error-prone, and fundamentally incompatible with the real-time data demands of industrial AI.
What happens at each layer
Understanding edge to enterprise requires understanding what each layer does and what each layer requires from the layers below it. The failures that prevent OT data from reaching enterprise systems almost always occur at one of these transition points.
How the SWTB product stack covers the full journey
SWTB builds the connectivity, contextualization, security boundary, and routing software that implements the edge-to-enterprise architecture across all five layers. The stack does not require replacing any existing control infrastructure: it adds the data pipeline layer above existing PLCs, DCS systems, and SCADA platforms.
What happens to data as it travels the pipeline
Data does not simply move from Layer 1 to Layer 5 unchanged. Each layer transformation is substantive: the data acquires context, changes format, changes granularity, and changes cardinality at each step. Understanding these transformations clarifies why the pipeline has the structure it has and what goes wrong when any transformation is skipped.
| Layer transition | What changes | Protocol / format | What breaks if skipped |
|---|---|---|---|
| Field → Edge (L1 → L2) | Multiple proprietary protocols unified to OPC UA. Raw register values become named, typed process variables with engineering units and quality fields. | Modbus/EtherNet/IP → OPC UA | Every downstream application must implement its own driver for every device protocol, creating N×M integration complexity across N applications and M device types. |
| OPC UA → UNS (L2 internal) | Flat OPC UA tag addresses acquire ISA-95 equipment hierarchy context. Topics become self-describing. Report-by-exception filtering reduces volume 80%. Edge analytics results are co-published. | OPC UA → MQTT Sparkplug B | Downstream consumers cannot identify which asset a reading belongs to. AI models cannot route readings to the correct model. Every consumer must maintain its own tag-to-asset mapping table. |
| OT network → IT network (L2/L3 → L4) | OPC quality fields are preserved through the boundary. Store-and-forward fills gaps. Outbound-only connection architecture prevents IT-to-OT inbound access. Write operations are explicitly authorized. | OPC tunnel / MQTT TLS | Security boundary is either bypassed (OT network exposed) or data quality fields are stripped (consumers receive stale data with no indication of its quality status). Network disruptions create data gaps. |
| IT-side OPC → Historian / MES (L4 internal) | Real-time process values become time-stamped archive records. Events become work orders and production records. Data is routed to the correct destination system based on event type and content. | OPC DA/HDA → Historian DB | REST/SQL → MES | Historian archive gaps prevent predictive AI model training. MES never receives production event data. Maintenance work orders are not auto-created from predictive AI alerts. |
| Plant systems → Enterprise / cloud (L4 → L5) | Plant-level data is aggregated to multi-site view. Production records are integrated with ERP order and inventory data. ISA-95 context enables cross-site benchmarking. Cloud compute applies ML inference. | REST API | SQL | cloud connectors | ERP operates on manual production data entry with hours to days of latency. Cloud AI has no access to real-time OT data for inference. Multi-site visibility requires manual consolidation. |
Why edge to enterprise is the prerequisite for industrial AI
Industrial AI, predictive analytics, digital twins, and generative AI copilots are all data consumers. Their value is determined by the quality, completeness, and latency of the OT data supply chain that feeds them. An AI model for predictive maintenance that cannot access real-time vibration data from the assets it monitors cannot produce real-time predictions.
For a high-speed production line or a compressor surge control application, a 2-second round-trip to a cloud server is an eternity. Edge compute processes control logic in milliseconds locally. The cloud layer receives filtered, contextualized summaries, not raw telemetry streams. Report-by-exception ensures only actionable changes traverse the pipeline.
The iDMZ outbound-only architecture delivers both: OT systems have no inbound internet exposure, but enterprise systems receive real-time process data continuously. TLS encryption, certificate-based authentication, and outbound-only connection initiation keep the OT network locked down while the IT network gets full data access.
For multi-plant operations, edge-to-enterprise architecture enables a single enterprise view of production performance across all facilities. Each site deploys the same edge-to-enterprise stack, publishes to its site-level UNS, and routes to the enterprise cloud layer. A single Power BI or Snowflake dashboard can compare OEE across 20 sites using the same ISA-95 context structure at each one.
AI and ML systems need clean, labeled, contextualized, complete time-series data. The edge-to-enterprise pipeline provides exactly that: ISA-95 context labels every reading, OPC quality fields flag unreliable data, store-and-forward ensures no gaps in training archives, and report-by-exception reduces noise.
Edge compute at Layer 2 runs autonomously from the IT network and cloud. If the cloud connection goes offline, the edge layer continues operating: sensors are monitored, local control logic runs, anomaly detection continues, and data is buffered for transmission when connectivity is restored. Store-and-forward in DataHub ensures no archive gaps.
Edge-to-enterprise architecture does not require replacing existing equipment. TOP Server connects to PLCs from the 1990s alongside the latest IEC 61131-3 controllers via its protocol driver library. N3uron applies ISA-95 context regardless of the device's age or connectivity capability. A factory with 40-year-old Modbus RTU devices can participate in the same edge-to-enterprise pipeline as a facility with native OPC UA controllers.