I Almost Bought a $90,000 Transformer. What a Power Quality Analyzer Showed Us Instead.

October 2023, the bill came in

It was a Tuesday morning, 6:39 a.m., when someone in finance forwarded a message titled 'FYI – utilities trending up.' Attached was a spreadsheet summarizing the electric bills for our three sites. I handle facilities and equipment purchasing for a 340-person processing company — roughly $380,000 in annual orders across 14 vendors. The electric bills run through me too. So the spreadsheet landed in my inbox.

First number: total electric spend up 17% year over year. Second number: base energy rates were up only about 4%. Everything else was demand charges.

Quick explanation for anyone who doesn't sit next to these bills. A demand charge isn't based on how many kilowatt-hours you consume. It's based on the highest 15-minute window of power you pulled during the billing period. Ours had gone from about $4,100 a month to $5,700. According to EIA figures, the U.S. average industrial retail price was around 8.4 cents per kWh in 2024 — and the demand charge sits on top of that, which is where a facility can quietly lose a lot of money.

My first reaction was reflexive.

My first guess was the transformer. It was wrong

Our largest plant runs a 1000 kVA oil-immersed transformer that was installed in 2019. During production pushes it was already sitting near 85% load. So when peak demand started climbing, I assumed capacity was the story.

I told our operations lead we probably needed to step up to a bigger unit. He asked me to get a quote. So I did: a 1500 kVA dry-type transformer, installed, came back between $81,000 and $93,000 with a 12 to 16 week lead time.

Then the utility's field engineer called me back. One thing he said annoyed me, and he turned out to be right: 'Don't buy a transformer until you know what the load curve looks like.'

How does a power quality analyzer actually work (it took me a week to understand this)

Before that call, my mental model of a power quality analyzer was 'a more expensive meter.' That's not it.

A normal meter reads at intervals — every 15 to 60 seconds, depending on the setup. A power quality analyzer samples thousands of times per cycle. What it catches:

• Peak demand in 15-minute windows, down to the kW
• Voltage sags and swells that last only a few cycles
• Voltage imbalance between phases
• Power factor — basically how much money you're paying for current that isn't doing useful work
• Harmonic distortion, which is how much non-linear load is dirtying the waveform

It doesn't fix anything. It measures.

That's the uncomfortable part, because the first time you see your own plant's load curve, the data usually doesn't point to a clean one-thing-to-buy answer.

Here's the reference point you need for the harmonics part. Per IEEE Std 519, total harmonic distortion at the point of common coupling should stay under 5% for most industrial facilities.

We rented an analyzer for 90 days. About $3,800 including installation and software.

Day four, I nearly worked off bad data

This is where the story turns.

We mounted the analyzer on the load side of the switchgear instead of the line side. Sounds like a small detail. It meant the unit was reading the combined current of four feeder circuits instead of the incoming service. For four days, the data was garbage.

We caught it only because the reported peak demand came back higher than the utility's own meter reading — which shouldn't happen. Called the contractor, re-mounted on the line side, started over.

Once the data rebuilt, here's what we were actually looking at:

• Peak demand at 780 kW, clustered between 2:05 and 2:20 p.m.
• Voltage THD between 7.1% and 7.6%, above the 5% threshold in IEEE 519
• Power factor at 0.82, which meant we were paying a penalty for reactive power

The actual problem wasn't transformer capacity. The transformer was fine. The problem was a 300-ton press and two air compressors starting inside the same afternoon window, plus harmonics making it worse.

So yes — I nearly spent $90,000 to solve a problem a $3,800 study eliminated.

The three weeks: bigger transformer, or a system

Once I had data, I did the thing I always do. I went back and forth. For about three weeks.

Option A was the 1500 kVA dry-type unit. $81,000 to $93,000, 12 to 16 weeks. It was budgetable, and honestly it was the path of least resistance — nobody questions a bigger transformer. But it would've increased headroom without touching the harmonics issue or the demand charge. It just would've added margin to an assumption I shouldn't have made.

Option B was keeping the existing 1000 kVA transformer and adding monitoring plus a demand response layer — a power quality analyzer, smart metering, a demand response management system, and an active harmonic filter. Roughly $31,000 all in, about six weeks, but it required a person to actually watch a dashboard. That meant a different kind of cost.

I chose B. The transformer wasn't the problem, and buying a bigger one would've been paying to confirm my own guess.

For context on why B was even an option: per FERC's annual demand response assessments, enrolled demand response capacity in the U.S. has run above 30 GW. That volume exists because demand response does one specific job well — it shaves the peak so you don't have to buy oversized equipment.

Why GridSense, and when I'd tell you not to bother

GridSense made the shortlist for a practical reason. They do transformers — dry-type, oil-immersed, and custom builds — and they also do a demand response management system. One vendor instead of three, and one set of people responsible when something goes sideways.

What actually sold me was the measurement piece. The GridSense demand response management system is built around smart energy meter data rather than a monthly bill. I asked roughly three times how the thing worked, and each time I got a straight answer instead of a sales deck.

But I'm not going to pretend this fits everyone:

If you've got one site and three big loads, a standalone power quality analyzer with logging plus someone re-sequencing start times by hand will probably get you most of what we got. We did exactly that at our second site and skipped the full system.

If there's no meaningful demand charge in your rate structure, a demand response management system is money on a shelf. It exists to shave peaks. That's it. It won't fix a bad power factor or a harmonics problem on its own.

If your transformer genuinely is old and undersized, then a transformer quote and a monitoring quote from the same supplier — GridSense included — can be worth comparing head to head. That just wasn't our situation.

What the bill looked like afterward

Demand charges dropped about 19%. The active harmonic filter plus re-sequencing brought measured THD down to 4.6% at the service entrance, under the IEEE 519 threshold. Across all three sites, we're running $12,000 to $13,000 lower annually on electric spend.

The system cost about $31,000. Payback landed somewhere around two and a half years.

What I got wrong, and when I'd do it differently

It took me about two years and three separate equipment purchases to understand that most facilities don't have a power problem. They have a measurement problem. If you don't know where the peak comes from, every solution you buy is a guess with a purchase order attached.

Looking back, I should've run the 30-day power quality study before I ever asked for a transformer quote. At the time I was under pressure to show movement, and a bigger transformer looked more like movement than four weeks of measuring. That instinct hasn't fully left me.

If you're staring at an electric bill you can't explain, the order matters more than the purchase: measure what you have before you decide what you need. The downside is that the answer usually isn't 'buy this one thing.' The upside is you won't spend $90,000 solving a problem you invented.