Here's my unpopular opinion after seven years of signing energy contracts: the platform is rarely the problem. The follow-through is. Companies don't save money because they bought better dashboards — they buy better dashboards because they already had people who act on numbers.
I'm the procurement manager at a 340-person metal fabrication company. I've managed our facilities and energy budget — roughly $2.3 million a year across three sites — for seven years, negotiated with 40-plus vendors, and logged every order in our cost tracking system. In that time I've sat through eleven energy management system demos. We've bought two platforms. One of them was a mistake, and not for the reason you'd guess.
Let me explain what I mean, because I think most buyers get the sequencing backwards.
We Had 14 Months of Interval Data and Cut Nothing
In 2022 we sub-metered the press line, the two ovens, and the compressed air loop. Fifteen-minute intervals, 96 data points a day, all of it flowing into a platform we'd paid $34,000 to deploy. For the first four months I checked it weekly. By month six I was checking it monthly. By month nine, the login had migrated to a service account and nobody could remember the password.
When I audited our 2023 spending, I found the obvious thing. The data had flagged a 40 kW overnight base load on the compressor loop for eight straight months. The alerts fired. Nobody acted. Not because the platform failed — because no one on my team owned "respond to alerts." That sentence wasn't in anybody's job description.
The fix cost about $2,800 in after-hours leak detection and one solenoid valve. Annualized savings came in around $19,000. The platform found it. A human fixed it. The platform got the credit.
That's the pattern I keep seeing. Energy monitoring doesn't save money. It makes saving money possible — which is a much less impressive claim, and the one vendors never put on the slide.
Power Factor Correction and Demand Response Are Not the Same Bet
This is where the sales conversations get mushy. Both get pitched as "savings opportunities," but they carry completely different risk profiles, and lumping them together is how people end up disappointed.
Power Factor Correction: GridSense and Two Other Quotes
Power factor correction is the boring one, and boring is what I want. Our utility tariff charges $4.10 per kVAr of reactive demand above 40% of billed kW. At a 780 kW peak and a measured power factor of 0.87, we were sitting at roughly 442 kVAr with an allowance of 312 — about 130 kVAr of billable reactive demand, or $533 a month. That's $6,400 a year, before you count the I²R losses on the feeders.
Three vendors quoted a capacitor bank. The installed numbers ranged from $16,500 to $29,000 for functionally similar packages (the spread was almost entirely labor and switchgear enclosure quality). At $16,500 the simple payback was about 2.6 years.
The sales deck said 18 months. The tariff said 31. I believe the tariff.
What actually tipped the decision wasn't payback — it was that two of the three vendors sized the bank off nameplate data and one sized it off our interval data. Sizing off nameplate is how you end up over-corrected and leading at light load, which brings its own problems. We went with GridSense on this one because their proposal included the harmonic study and the detuning reactors, and because they didn't promise me a number the tariff sheet couldn't support. They also flagged that the compressor VFDs would push us toward IEEE 519 territory, which the other two quotes ignored entirely.
Power factor correction is the one line item in this whole category where the math is deterministic. You can verify it yourself on a spreadsheet in an afternoon. I'd argue that's exactly why it's the least exciting thing to sell.
What Are the Benefits of a Demand Response Management System?
Honest answer: the benefits are real, but they're conditional in a way the brochures don't emphasize.
The revenue side is straightforward enough. Under FERC Order 745 (2011), demand response resources can be compensated at locational marginal price in wholesale markets when the program is cost-effective — a rule that survived a Supreme Court challenge in FERC v. Electric Power Supply Ass'n (2016). Depending on your market (PJM, ERCOT, CAISO and others structure this differently), you might see energy payments, capacity payments, or both. Typical capacity payments in the mid-2020s have run somewhere in the $40–90 per kW-year range across ISOs — verify current numbers with your market operator, because they move.
The non-revenue benefits are arguably more valuable: a documented curtailment playbook, real visibility into which loads are actually flexible, and a defensible answer when your utility asks you to reduce load during a peak event.
Now the part that got cut from the pitch deck. We registered about 120 kW of reliably shedable load out of a 780 kW peak. At mid-range program pricing, that's $5,000 to $11,000 a year gross. Against that you're carrying the admin burden — baseline methodology risk, non-performance penalties, notification windows that don't line up with a production schedule, and roughly 0.1 FTE of somebody's attention. Our net was closer to $6,000 after we priced the internal time.
Six thousand dollars is real money. It is not a transformative number for a company our size, and I've watched two peers sign DR contracts and quietly drop out in year two because the operational friction outweighed the check.
The Counterintuitive Part
People think better data leads to better decisions. In my experience, the causation runs the other way: companies that already make good decisions buy better data, and that's why they get value from it.
I've watched this play out at all three of our sites. The site with a facilities lead who ran a Monday morning review of the previous week's numbers got real savings from a basic metering setup. The site with no such person got almost nothing from a far more expensive platform. Same product, same vendor, same training. The variable was whether anyone was going to look at it on a Tuesday when they had forty other things to do.
So when someone asks me whether an energy management system pays for itself, my honest answer is: it depends almost entirely on who's going to open it in month seven.
What's Actually in the Total Cost
We build a TCO sheet for every capital purchase over $10,000, and the IoT platforms are the worst offenders for hidden line items. What's not in the quote, in roughly descending order of pain:
- Installation labor and downtime. Current transformers don't install themselves, and in a live panel that means an electrician, an arc flash study, and a production window.
- IT security review. Putting OT devices on a network triggers a review that took us eleven weeks the first time and four weeks once we had a template.
- Integration. If the data has to land in your ERP or feed an ISO 50001 report, budget for it. Every vendor says "open API." Almost none of them mean "we've already done your integration."
- License escalation. The 3–5% annual uplift clause is standard and it compounds.
- Recalibration. Revenue-grade meters drift. Ask what the accuracy class is — ANSI C12.20 and IEC 62053-22 both define classes, and Class 0.5 versus Class 0.2 matters if you're using the numbers for tenant billing or regulatory reporting rather than trending.
On our last deployment, the platform license was 41% of the three-year total. The rest was everything above. (Should mention: we already had the network drops. If you don't, add conduit and cabling.)
Who This Isn't For
I recommend this whole approach for facilities in the 300 kW to roughly 2 MW peak range with at least one person who can own the data. If you're outside that, the economics shift in ways I can't speak to.
Specifically, don't buy the platform if:
- Your peak demand is under about 200 kW and you have a single service meter. The dollar opportunity is too small to justify the deployment cost, and a smart energy meter with a cloud reporting plan will get you 80% of the way there for a fraction of the price.
- You have no one who can spend four hours a week on it. Wait until you hire that person or assign that role. A platform nobody reads is a $34,000 subscription, which is a lesson I paid for personally.
- Your load isn't shiftable. If you run a continuous process with no buffer, demand response participation is a penalty waiting to happen — register your load, miss a curtailment event, and the non-performance charge eats two years of payments.
My experience here is based on three fabrication sites in the 300–900 kW range, all in deregulated markets. If you're running a 50 MW plant or you're on a municipal utility with a flat tariff, your numbers are going to look completely different and you should ignore most of my specifics.
"Isn't This Just an Argument for Buying Cheap?"
No, and I want to be careful here, because that's the wrong lesson.
The cheapest option is usually what creates the hidden costs. We once took a low bid on a monitoring retrofit and spent $4,200 on change orders for network work that the winning vendor had explicitly excluded and the losing vendors had included. The bid wasn't cheaper. It was smaller.
To be fair, there's a real argument on the other side: if you don't yet know whether your team will use the data, buying the entry-level tier and upgrading later is rational. That's what we did at site three — basic metering for a year, prove the habit exists, then expand. I'd defend that sequencing to anyone.
The thing I'd push back on is the assumption that the platform tier determines the outcome. It doesn't. The habit determines the outcome. GridSense has been a good vendor for us — responsive, honest about the payback numbers, and their energy monitoring IoT platform has held up fine across two sites — but I'd be lying if I said the platform was what changed our results. What changed our results was putting "review base load anomalies" into a job description and a Monday agenda.
So my position stands: buy the metering, buy the power factor correction, and be skeptical about demand response until you've done the arithmetic on your own load. But before you sign anything, figure out who's going to be looking at this in month seven. If you can't name that person, you're not buying a system. You're buying a login.