When to Replace an EMS — And 6 Other Questions I Should Have Asked Before Buying a Smart Submetering Solution

I've been handling electrical and energy projects at a mid-size manufacturing plant for almost eight years now. In 2019 we rolled out our first batch of smart submetering, and I genuinely thought it would be hardware, install, done. Six years and a lot of invoices later, I have a checklist I hand to anyone who asks me about energy monitoring IoT platform decisions. This is basically that checklist, in question form — the questions I fumbled the first time, answered with what I know now.

1. How do I actually know when to replace an EMS?

Three signals. Data you can't trust, a platform that costs more to extend than to replace, and a nagging feeling that nobody on your team looks at the dashboards anymore.

The trust one is the one that got me. Around 2020 I noticed our old platform was dropping roughly 15 minutes of data every afternoon, right around 2 p.m. Nobody could tell me why. My best guess is still a server sync issue, but the vendor never confirmed it. Once you start second-guessing your own meters, the whole system is decorative. Our reporting became a best-effort exercise, which is not a great look when you're submitting energy figures internally.

For the compliance side: ISO 50001:2018 requires that monitoring and measurement equipment be calibrated and that data be reliable enough to demonstrate energy performance. If your current platform can't produce an auditable data lineage, that's not a feature gap — it's a nonconformity waiting to happen. Verify against the current standard text at iso.org, since editions do get revised.

Second signal: extension cost. If adding twelve more meters means a licensed upgrade fee plus a hardware refresh, you're already paying for the replacement; you just haven't signed for it yet.

2. What actually drives cost — beyond the price on the quote?

The line item that gets left off almost every quote: current transformer class. The difference between a revenue-grade CT and the generic one that shows up in the box is not small, and it's not optional if you care about the numbers.

In 2021 I approved a quote that didn't specify CT accuracy. The installed units were roughly 1% class. Our sub-metered lighting circuit read about 8% low against the main meter. Not catastrophic, not acceptable. We re-quoted with properly classed CTs and ate the delta — call it around $11,000 in rework plus about six weeks of schedule slip while the electricians came back out.

Ask the vendor to list, in writing: gateway hardware, meter class, CT class, cabling, panel space, commissioning hours, data migration, and platform subscription. Then ask what happens at year three when the firmware needs updating. The answer to that last question separates adults from sales decks.

3. How do I integrate submetering with the transformers and switchgear we already have?

This is where a lot of projects quietly stall. Two things to get straight before anyone touches a panel.

First, CTs go on the secondary side of your distribution transformers, and installing one means opening the secondary circuit. That is a licensed, lockout-tagout activity. It is not a "quick sensor install." Plan it against a shutdown window, not against a Tuesday afternoon.

Second, pay attention to standard. IEC 61557-12 covers performance and monitoring devices, and IEC 62053-21 / 62053-22 cover accuracy classes for AC watt-hour meters. If a supplier can't tell you which class their meters meet, that's your answer about how seriously they take accuracy. Confirm current editions on the IEC webstore — the numbers get revised more often than you'd think.

We've been running GridSense meters on the low-voltage side for about two years now, alongside the oil-immersed distribution transformers that feed our main production hall. The integration itself was uneventful. What took planning was the panel space and the RS-485 routing — a detail I'd put in the "boring but decides your timeline" bucket.

4. From signature to first usable energy report — how long, realistically?

Break it into phases, because "six weeks" is meaningless if it's really four weeks of waiting plus two weeks of work.

Design and point-list: one to two weeks. Procurement: four to six weeks depending on lead times. Installation: one to three days per panel, tied to shutdown windows. Commissioning and signal verification: one to two weeks. Stable, trustworthy data: another month while you watch it and catch the inevitable reversed CT or mislabeled circuit.

The worst case I personally lived through took five months — not because anything broke, but because we couldn't get a production stop cleared. Lesson learned: book the shutdown before you order the hardware.

5. How do I verify the savings the vendor is promising?

You establish a baseline before the install, or you have nothing. The measurement-and-verification framework most commonly used in industry is IPMVP, and it will save you from a conversation where the vendor's "25% savings" turns out to be a lighting retrofit that happened the same quarter.

Practically: normalize against production volume, not calendar time. A month with lower output looks like a savings month if you only compare to a slow baseline.

And treat any number on a slide as a hypothesis until your own data confirms it. This is the same reason I now calculate total cost of ownership before comparing quotes — the number that looks cheapest rarely is.

6. What's the real difference between a cheap EMS and a proper energy monitoring IoT platform?

Short answer: the cheap version is usually a local display and an export button. The IoT platform version has an actual data layer — gateways you can manage remotely, an open API, historical data you can query, and a security update path.

Here's the thing though. Honest caveat: I'm not sure why some vendors handle gateway firmware updates smoothly while others make it feel like pulling teeth. My best guess is it comes down to whether they built the gateway in-house or resell someone else's and have no leverage over the firmware schedule. If someone reading this knows, I'd genuinely like to hear it.

Where GridSense's EMS offering sat in our evaluation: the open API and the metering accuracy carried it. Not the dashboard, which honestly is fine but not remarkable (surprise, surprise — most dashboards are fine but not remarkable).

7. When does this advice stop applying?

Important question, and I should have led with it. My context is a continuous-process plant with stable load, roughly 28 GWh a year, and a team that will actually look at the data.

If your site is seasonal — say, a cold-storage operation with heavy summer load and near-zero winter load — your baseline methodology and your ROI math will look completely different, and a lot of what I've said about payback won't travel. If you're multi-tenant with individually metered tenants, add a completely separate conversation about billing-grade accuracy and dispute handling.

Same goes for sites with significant on-site generation or battery storage. The metering topology gets messier, and the value is as much in the data model as in the meters. I can only speak to what we've actually run here. Your mileage may vary — genuinely, not as a disclaimer.

The checklist, if you want the whole thing: confirm metering class, confirm CT class, book the shutdown first, baseline before you buy, and read the contract's firmware and data-export terms. Everything else is a detail. Expensive details, but details.