I've been a controls engineer for eight years, and for the first five of those, “it supports Ethernet” sounded like a complete answer to me. Then, in late 2019, I delivered a machine with a Siemens S7-1500 into a plant where every other controller was a Rockwell ControlLogix. The machine builder had promised the two systems would “talk to each other with no problem.” The plant engineer heard what he wanted to hear. Nobody wrote the words PROFINET or EtherNet/IP anywhere in the specification.
That omission cost the project somewhere around $12,000 in rework, travel, and change orders. It was the first of three integration mistakes I've made and documented, and together they've cost me and my customers roughly $40,000. I now maintain our team's pre-shipment communication checklist, and this article is basically the short version of why that checklist exists.
Spend any time in industrial automation and you'll run into the Allen Bradley vs Siemens PLC debate. Which CPU is faster? Which programming environment hurts less? Which brand has better support? I've enjoyed those arguments myself. They're also mostly irrelevant inside a real plant.
You rarely get to build a greenfield line and pick one brand for everything. You inherit a plant standard. The machine builder has a standard. The maintenance team knows one platform and resents the other. Spare parts, training budgets, and support contracts don't reset just because you bought something new.
So the practical question isn't “which brand should I buy?” It's “these two CPUs already live in the same building—how do they share data without me rebuilding one of them?” Most buyers focus on choosing the “better” PLC and completely miss the interface between the PLCs they already own. The question everyone asks when they spec a new line is “Allen Bradley or Siemens PLC?” The question they should ask is “what will this machine talk to, and who proves the data map before shipment?”
Here's the trap: PROFINET and EtherNet/IP both run on standard Ethernet. Same cables, same switches, same RJ45 connectors. If you stare at the wire, there is no visible difference. But they are different protocols, and neither controller speaks the other's natively.
PROFINET is maintained by PROFIBUS & PROFINET International and is standardized under IEC 61158. EtherNet/IP is ODVA's version of the Common Industrial Protocol—CIP—running over Ethernet. Both ecosystems are capable. They're just not the same language. On the equipment I work with daily—S7-1500s configured in TIA Portal and Logix CPUs in Studio 5000—the Siemens CPU won't present itself as an EtherNet/IP device to a Rockwell scanner, and a ControlLogix won't show up as PROFINET IO to the S7 side. There's no checkbox to enable. You need something in the middle.
This is not a conspiracy. Both companies built their architectures around their own ecosystems, which is their right. But if you show up at a plant with one of each, the outcome is the same: the devices can share a network, but they will not discover each other, exchange I/O, or share tags until someone builds a translation layer.
Even after you add a translator, a sneakier problem appears: the data itself. Siemens S7 data is transported big-endian—most significant byte first—while a Rockwell ControlLogix expects little-endian values. A 32-bit REAL that means 4.5 bar in the S7 can arrive at the Logix as a number in the 10^-38 range if the byte order is left at default.
I know that specific number because I watched it happen in September 2022. We had a pressure transmitter reading 4.5 bar. The tag on the ControlLogix side showed 8.2e-38. Every diagnostic light on the gateway was green. The network was healthy. The protocol conversion was working exactly as configured. The data was just nonsense because nobody had configured byte order. A colleague had flagged it during the internal review: “did you check the byte order on the REALs?” I remember answering, “the gateway handles that.” It does not. A gateway converts protocols—it does not know that 4.5 is a pressure, a distance, or a torque. Data type and byte order choices are application decisions, and I'd left them to a default.
One setting on the mapping table fixed the value. The invoice for that site trip was not nearly as painless. I only fully believed in data mapping validation after ignoring a warning in exactly that way, and that warning is now part of every kickoff meeting: if two PLC brands share data, someone must own the byte order.
The 2019 bill looked like this:
I can't give you a clean total, because I stopped adding in week three. North of $12,000, before anyone counted lost production. What I remember more clearly is how the conversation changed afterward. Every review meeting started with “with all due respect, the machine builder said this would be plug and play.”
The damage isn't only the dollar figure. On a big site with a dedicated controls team, you can absorb a mistake like that. On a smaller plant, where one or two people keep everything running, an interface failure steals their entire week. They don't have an integrator war room or a spare engineer. They have a line that won't run and a quote that didn't include this.
I don't think big automation customers and small ones get treated differently on purpose. It happens because small projects get less documentation time, and documentation time is exactly where integration mistakes are caught.
After the September 2022 trip, I wrote our team's checklist for any project where a Siemens PLC needs to talk to a Rockwell controller. It's deliberately short:
The most reliable routes are usually one of three choices—a hardware gateway for cyclic I/O between two CPUs, an OPC UA bridge when the data goes into a higher-level system anyway, or open TCP communication when the data is small and there is no budget for extra hardware. Choose one before you write the PO, not during commissioning.
I'll also say this for anyone who thinks their project is too small to need the full checklist: it isn't. The same process that costs one extra engineering day before shipment prevents two weeks of site firefighting. Small manufacturers often feel like integrators don't take them seriously, and I understand why. But a $4,800 integration done properly is how a customer decides you're the person they call for the $400,000 line replacement. Small doesn't mean unimportant. It means potential.
By January 2024, after one near-miss that I don't want to repeat, the checklist became mandatory. Since then, we've caught 31 potential integration errors across 19 cross-vendor projects. None of them was a broken wire. Each one would have shown up on-site, at site prices, with the customer watching.
So when someone says “Siemens and Rockwell PLCs talk to each other all the time,” they're not wrong. They're just leaving out the part where somebody has to decide exactly how. Write down who that somebody is, what protocol sits in the middle, and what the data map says—before the hardware ships.
It's cheaper than the version I paid for.