How to Bridge the Gap Between Hardware and Digital Workflow

Shubham Kumar

16 Jun 2026 — 7 min read

Factory floors and software systems still don't speak the same language. Machines produce data. Platforms need data. And somewhere between the CNC machine and the ERP dashboard, something disappears. This piece looks at what that gap actually is, why it keeps showing up even in well-funded operations, which technologies are making a difference — and which ones are still catching up.

The Gap Is More Specific Than You Think

Most people assume this is a technical problem. Wrong protocol, wrong format, wrong API. Turns out, it's usually something simpler and more frustrating — organizational. The machine side and the software side are owned by different teams with different priorities. The maintenance engineer doesn't care about cloud architecture. The IT manager doesn't understand why the legacy PLC can't just "send JSON."

That's where purpose-built approaches matter. Providers focused on manufacturing IT services, DXC Technology among them, have been developing structured integration frameworks designed to handle both sides simultaneously — legacy equipment and modern data infrastructure — without forcing manufacturers to rip out perfectly functional hardware worth millions.

The result? You don't replace the machine. You build a translator between it and everything else.

Sounds simple. Takes months to get right.

What the Market Actually Looks Like Right Now

Technologies That Actually Shipped

A few concrete things changed in the last two years that made this conversation more practical:

OPC-UA became the de facto standard for industrial device communication. Siemens, Bosch, and Rockwell Automation pushed it hard across product lines. If your facility runs modern PLCs, there's a decent chance OPC-UA endpoints are already sitting there, unused.
Edge computing stopped being a conference talking point. AWS Outposts, Azure Stack Edge, and comparable products physically sit on factory floors now, processing data locally before anything reaches the cloud. This matters when machines produce 50GB of sensor data per hour and cloud egress fees would make the whole thing economically nonsensical.
Digital twin platforms moved from demo stage to actual production use. PTC's Vuforia, Siemens' Xcelerator, NVIDIA's Omniverse — BMW runs digital twins of entire assembly lines. So does Airbus. These aren't concept videos anymore. They're operational infrastructure.

What's Still in Prototype Phase

Not everything is production-ready. Some of the more interesting experiments running right now:

AI-based anomaly detection on raw machine signals. Augury and Samsara train models directly on vibration and acoustic data — catching motor failures before anything lights up on a traditional monitoring screen. Results in controlled deployments look solid. Broad rollout? Still limited. Most facilities aren't ready to trust a neural network over a maintenance engineer with twenty years on the floor.
Natural language interfaces for SCADA. Operators asking machines questions in plain English instead of navigating multi-layer HMI menus. Honeywell had a working prototype at Hannover Messe 2024. Genuinely impressive in a demo setting. Whether it holds up at 3am during a shift handover, with a stressed operator and a partially-failed sensor array — that's the test nobody's published results on yet.
Autonomous mobile robots with dynamic path planning. Not the pre-programmed track-following bots. Systems that reroute in real time based on floor conditions. Geek+ and 6 River Systems (the one Shopify acquired, then eventually let go) are deep in this. Warehouse environments first, shop floors next. The hardware works. Integrating it with existing WMS and MES without creating scheduling chaos is where projects stall.

Why Hardware Still Wins Arguments

A well-maintained CNC from 2003 can hold tighter tolerances than some newer models. Proven conveyor systems don't need firmware updates to keep running. When someone says "modernize," they rarely mean replace the machines. They mean the data those machines produce is stranded — and the software layer needs to catch up.

Most facilities have three layers where this becomes a problem:

Field level — sensors, actuators, PLCs. Data here lives in proprietary formats: Modbus, PROFIBUS, OEM protocols nobody outside that vendor fully documents. Getting clean data out is step one. Usually the hardest.
Control level — SCADA, DCS, MES. Some talk to field devices. Few talk to each other.
Enterprise level — SAP, Oracle, Dynamics. Almost always completely disconnected from what's happening on the floor in real time.

Bridging all three without halting production is the actual work. No shortcuts. No single platform that solves all three at once, regardless of what the sales deck says.

Practical Integration

The most reliable approach right now is edge gateways with protocol translation built in. Devices like Kepware's KEPServerEX (a PTC product), Moxa's industrial edge computers, or Advantech's WISE series sit between old machines and modern networks. They speak the legacy languages — Modbus RTU, PROFIBUS DP, EtherNet/IP — and output structured data that modern systems can actually consume.

Setup isn't trivial. A Kepware deployment for a mid-sized facility might take two to three weeks of configuration, tag mapping, and testing. But that's a fraction of the cost and disruption of replacing the machines themselves.

What Good Data Architecture Looks Like

Once the data is flowing, architecture decisions start to matter:

Time-series databases — InfluxDB, TimescaleDB, OSIsoft PI (now AVEVA PI). These were built for machine data. Relational databases can store it, technically. But they weren't designed for 10,000 data points per second from 400 sensors.
Message brokers — MQTT is now everywhere in manufacturing, originally built for constrained IoT environments. Apache Kafka for high-throughput scenarios where you genuinely can't afford to lose a message. RabbitMQ for lighter, more transactional workloads.
Raw storage vs. operational stores — raw data lands somewhere like Azure Data Lake or AWS S3. Processed, clean data goes into systems that ERP or MES platforms can actually query without choking.

The architecture doesn't need to be elegant. It needs to be intentional. An undocumented pipeline is just future technical debt with extra steps.

The MES Problem Nobody Wants to Admit

Manufacturing Execution Systems are supposed to be the bridge between floor and enterprise. In theory. In practice, most MES deployments are five to ten years old, customized beyond recognition, and partially owned by someone who left the company three years ago. SAP ME, Siemens Opcenter, Rockwell's FactoryTalk — all solid platforms that turn into nightmares when customizations go undocumented for a decade.

The pragmatic move? Don't try to fix the MES first. Build the data pipeline around it. Make the MES a consumer of structured data. Clean it up once the pipeline is stable. Trying to modernize the MES and the connectivity layer at the same time is how integration projects stall for eighteen months and burn through budgets.

The Human Factor — Which Is Most of the Problem

Talk to anyone who's actually deployed these systems, and the story rhymes: the technology held up. The people didn't.

Operators who've run machines for fifteen or twenty years don't trust dashboards they've never seen before. Maintenance teams don't report sensor readings as data — they report them as "the machine sounds off." IT teams apply corporate network security models to OT environments, creating vulnerabilities in both directions. OT teams treat IT involvement like a threat to operational continuity. Which, given some project histories, is understandable.

None of this is a technology problem.

Some approaches that actually help:

Co-location during rollout — OT and IT people in the same room for the first few months. Not coordinating across departments. Actually sharing a physical space and a whiteboard.
Early wins that operators care about — find one metric that matters on the floor (uptime, scrap rate, first-pass yield) and demonstrate improvement within the first 90 days. Credibility opens doors that project charters don't.
Hands-on training — not slide decks, not e-learning modules with completion checkboxes. Actual operators in front of actual dashboards with real data, allowed to make mistakes in a test environment.
Operator feedback mechanisms — production floor workers often know things the sensors don't. If a dashboard shows normal and the operator says something's wrong, the operator is usually right. Build that input into the system.

Real Deployments Worth Studying

Volkswagen's Industrial Cloud — Azure and SAP, 124 manufacturing facilities, project started in 2019 and still expanding. The goal: a process improvement found in one plant deploys across all 124 within days. That's the kind of scale where connectivity architecture stops being an IT discussion and becomes a competitive one.
Siemens Amberg — held up constantly as proof that full digitization is achievable. What the case studies leave out: the facility opened in 1989 and evolved through every generation of industrial technology over thirty-plus years. No single transformation event. Just incremental decisions, compounding over decades. Worth remembering before anyone pitches a two-year roadmap to the same outcome.
NVIDIA Omniverse in automotive — BMW and Mercedes-Benz both use it to simulate floor layouts before physically moving equipment. The ROI isn't in the simulation software. It's in avoiding the two to three weeks of production downtime that come from rearranging a line, discovering a new bottleneck, and having to do it again.
Rockwell and Plex Systems — Rockwell acquired Plex in 2021 specifically because hardware automation expertise alone wasn't enough. The integration is ongoing. But the acquisition itself is a public admission that the industry's old divisions — "we do machines, someone else does software" — stopped making sense.

Where This Is Actually Going

The honest direction: more standardization and less heroics per deployment.

The ISA-95 standard, which defines how enterprise and control systems communicate, is being updated to accommodate cloud-native architectures. The OPC Foundation keeps expanding OPC-UA scope. MQTT Sparkplug B is gaining adoption as a standardized payload format for industrial IoT — which sounds dry but practically means a sensor from Vendor A and an analytics platform from Vendor B are increasingly likely to work together without a custom integration project.

None of that is exciting to write about. But it's genuinely useful.

What's also changing: tooling costs. Industrial edge devices that ran $12,000–15,000 per node a few years ago are available under $1,500 now. Cloud analytics platforms ship pre-built connectors for most major PLC brands. The activation energy for starting is lower than it was in 2020.

The companies that struggle aren't the ones with old machines. They're the ones with old mental models about what IT is supposed to do on a production floor — and who owns what when something breaks.

Closing Out

Bridging hardware and digital workflow isn't one project with a launch date. It's a series of smaller ones — each making the data a little more accessible, a little more reliable, a little more connected to the decisions that matter operationally. The gap doesn't close in a quarter. It closes incrementally, messily, and almost never according to the original project plan.

The facilities getting this right aren't necessarily running the newest equipment. They're the ones that stopped waiting for a perfect solution and started building practical ones — with the machines they already have and the teams they already know.

That's usually enough to get started.