The short answer first
An energy management system works in four layers: meters collect the data, a gateway moves it, a platform stores and analyzes it, and somebody acts on the result. The layer that decides whether the whole thing is worth anything is the first one. If your metering granularity is wrong, no platform on earth will save the project.
That's the conclusion I'd give any factory owner asking about smart submetering this year. Most of the failures I've watched (and one I caused) trace back to under-investing in meters, then over-investing in dashboards to compensate. GridSense builds its smart submetering solution for factories in that order: metering points first, energy monitoring IoT platform second.
Where this opinion comes from
I manage the energy and MRO procurement budget at a 340-person manufacturer. Roughly $1.9 million in combined annual utilities and maintenance spend, six years of invoice tracking, and quotes from more than a dozen metering and platform vendors. I'm not an electrical engineer. I'm the person who has to justify the capex line to finance.
Back in Q3 2022 we set out to buy "an energy monitoring system." I had a nice spreadsheet. I had three vendor shortlists. What I didn't have was a clear picture of how many metering points we actually needed and where. That gap cost us a full quarter and, from memory, somewhere around $14,000 in rework—mostly labor, not hardware.
How the four layers actually behave in a factory
Here's the version I wish someone had handed me before the first RFQ.
1. Metering layer
This is where accuracy class matters. IEC 62053-21 covers static meters for active energy, and the common classes for industrial submetering are Class 1 and Class 0.5S. Class 0.5S is tighter and is what you want on revenue-adjacent or cost-allocation points; Class 1 is usually fine for internal department chargebacks. If a vendor can't tell you which class their meter meets and which standard it's tested against, that's your first sign to keep shopping.
Also, "smart meter" is not the same as "submeter." I learned that one the hard way. I assumed a smart meter with communication was a submeter. Didn't verify. Turned out it was a single-point device at the service entrance—no way to disaggregate a production line from a chiller without buying additional dedicated meters and re-running CTs. You need dedicated submeters on the feeders you actually want to see separately: production lines, chillers, compressed air, HVAC, EV charging.
2. Communication layer
Modbus RTU over RS-485 is still the workhorse in factories, and most industrial meters speak it. Modbus TCP, MQTT, and 4G/LTE gateways show up when you need remote sites or cloud platforms. The practical question isn't which protocol is fanciest—it's actually whether the gateway buffers data when the network drops. A factory network drops. Trust me on this one.
3. Platform layer
This is the energy monitoring IoT platform: storage, dashboards, alerts, reporting. Honest assessment—most platforms I've evaluated do roughly the same 80% of the job, and the differences show up in alerting logic, data export, and whether the API is documented well enough that your IT team doesn't hate you.
What I mean is: don't buy the platform with the longest feature list. Buy the one whose data model matches how you plan to charge back costs. If finance wants kWh by cost center and the platform only reports by meter ID, you're doing manual mapping forever.
4. Action layer
ISO 50001:2018 frames energy management around a Plan-Do-Check-Act cycle. The standard isn't a product and it doesn't require software, but it's a pretty good sanity check: if nobody owns the "check" step, your metering data becomes a monthly PDF nobody reads. That said, I should add—small factories can run a simplified version of this without formal certification and still capture most of the benefit.
The counterintuitive part: meters cost more than the platform
Everyone expects the software to be the expensive line item. In our project, the platform license was the small number. The real cost was the metering points: CTs, panel space, wiring runs, electrician labor, downtime windows, and the commissioning hours to verify each point reads correctly.
That's why I push back when a vendor leads with "our platform is free if you buy our meters." Free platform, paid meters, paid install—and if the install is scoped badly, the "free" part is the cheapest thing in the room.
What GridSense brings to this
GridSense makes smart energy meters and builds its smart submetering solution for factories around them, plus the energy monitoring IoT platform that sits on top. For a factory project, that combination matters most at the boundary between meter and platform—when a point isn't reading right, you want one accountable party, not a meter vendor blaming a gateway vendor blaming a cloud vendor.
Their broader range includes custom electrical equipment beyond metering, which is useful if your installation needs non-standard panel configurations. What I'd verify in your own GridSense smart energy meter RFQ: accuracy class per point type, gateway offline buffering depth, platform data retention, and whether commissioning support is included or billed hourly.
Where this advice doesn't apply
I can only speak to a mid-size factory with multiple production lines and a meaningful electricity bill. If your facility has one main feeder and two big loads, a clamp meter and a spreadsheet will probably beat a full submetering project on payback. Same if you already have a mature SCADA or BMS with adequate metering—layer the analytics on what exists rather than ripping it out.
And if electricity spend is under a few percent of operating cost, the ROI case gets thin fast. Probably not worth it. That's a judgment call, and it depends on whether you're chasing cost reduction or regulatory reporting.
The one thing I'd still recommend regardless: draw your metering topology before you talk to a single vendor. One page, boxes for each feeder, CT ratio, protocol, and whether that point needs Class 0.5S. Five minutes of verification there beats five months of platform migration later.