What the factory floor teaches you

There's a particular feeling, the first time a thirty-year-old PLC speaks fluently to a modern data layer. The signal comes through. The dashboard updates. The data engineers go quiet for a moment, then start asking questions you didn't expect.

This is a short note on what factory connectivity actually delivers — not the marketing version, the working version. We've spent recent work inside an MDR-compliant manufacturing environment, taking legacy machinery onto a Unified Namespace, building the dashboards and the pipelines that put production data in front of the people who need it. Some of what we learned was technical. The interesting parts were not.

Connection is the boring part

The work that takes the longest — months of it — is the unglamorous part. Identifying which signals matter. Translating between protocols that don't speak the same dialect of value. Naming things so the people upstream can find them. The Unified Namespace pattern helps; it's the lingua franca everyone has been waiting for. But the schema work is still real work, and skipping it produces dashboards no one looks at.

The day the signals start flowing isn't the win. It's the start.

Connecting things that weren't meant to be connected

A lot of the equipment on a regulated production floor was designed before networking was the default. Connecting it requires care — and the care is not mostly about cybersecurity, although that matters. It's the validation risk of touching qualified processes, and the operational risk of being the new thing that takes a line down.

The patterns we lean on, in roughly this order of priority:

• Read-only by default. Data flows out. Nothing flows back. The simplest possible architecture, and the one that earns trust from operators and quality teams within the first conversation.
• One-way gateways at the OT / IT boundary. Protocol gateways translate but don't expose. For the most critical interfaces, hardware data diodes.
• Segmentation that respects the existing topology. Cell-level VLANs, explicit allow-lists, no flat networks. The new system doesn't get to talk to anything it doesn't need to.
• Zero-impact installation. Any tap goes in alongside the existing wiring, not in series with it. Removable, off-switchable, observable. If the new system disappears tomorrow, the line keeps running today.
• Change documentation that slides into the existing validation system. A connection to a qualified machine is itself a change. We design and document it to fit the manufacturer's change-control process rather than fight it.

None of this is dramatic. All of it is the difference between a connectivity project the quality team trusts and one that takes a year to qualify.

The first thing you see is yourself

Real production data, surfaced in real time, holds up a mirror. Cycle times that were assumed to be steady drift more than expected. A machine that everyone "knew" was the bottleneck turns out not to be — there's something quieter upstream taking time no one was tracking. Operators who've been running a line for a decade see the line they've been running.

This is where the value starts. Not in the algorithms. In the visibility. The line manager asks for a chart they couldn't have asked for last quarter, because they didn't know it was possible.

The dashboard isn't the deliverable. The dashboard is what makes the next question askable.

Real-time efficiency is incremental, and that's fine

The headlines write themselves: AI! Manufacturing! Twenty percent gains! In practice, real-time data delivers two and three percent gains, repeatedly, in places nobody expected. A cooling cycle adjusted slightly. A material batch caught early. A scheduled maintenance brought forward by a week and saving an unplanned stoppage two months later.

None of these are heroic. Added up over a year, they're enormous. The discipline is to keep capturing them — and to keep building the systems that surface them. The single biggest predictor of whether a factory connectivity project pays off is whether the team treats those small gains as worth the work.

R&D inherits the data

The most interesting part comes later. After a year of data accumulating on a properly modelled namespace, the R&D function has a new input it didn't have before: the production record. Not a sample, not a summary — the whole thing.

That changes the questions R&D can ask. Where is the production variability we already absorb without noticing? Which design choices we made five years ago are now creating the most friction on the floor? What would the next product look like if we knew, in advance, the production behaviour we're going to inherit?

This is the long arc that factory connectivity makes possible. It's not "AI on the production line." It's the production line becoming feedback for what the company builds next. The factory teaches you things the lab can't.

What to do before you start

If you're scoping a factory connectivity project for a regulated environment, three things matter more than the technology choice:

1. Pick a single, valuable signal first. Not the whole shop floor. One machine, one signal, end to end, working. Everything else is easier after that.
2. Bring quality and regulatory in early. Production data in a regulated environment is not neutral — it's evidence. The systems that surface it have to be designed knowing that, from the first sketch.
3. Treat the dashboard team as engineers, not designers. The hard part of an operational dashboard isn't the visualisation. It's deciding which one of the eighty visible numbers is actually the question being asked.

We're writing more on this. If you're scoping something similar and want to talk it through, the contact form is the fastest way to start.