Smart Submetering, Demand Response, or Power Factor Correction First? A Scenario-Based Guide for Factory Energy Management

"Where should a factory start with energy management?" I've heard that question from two very different seats.

Four years on the buyer's side, reviewing vendor submissions and signing off on specs. Then the last few years at GridSense as quality and brand compliance manager, where I review roughly 300 technical documents a year before they reach customers. Same question, opposite direction.

The honest answer isn't a technology. It's a sequence. And the sequence depends on something most people skip: your utility rate structure.

It took me about 200 proposal reviews to accept that. The "best" answer kept flipping—sometimes toward metering, sometimes toward demand response, sometimes toward power factor correction—and the variable wasn't the plant size or the budget. It was the bill.

Three scenarios, three different starting points

Most of the factories I've reviewed fall into one of three buckets:

• New construction or major expansion — walls are open, decisions are cheap to change.
• Existing plant with painful demand charges — the load is already there, the question is control.
• Power factor penalties already on the bill — the utility has told you the number in dollars.

Each one points to a different first move. Get the order wrong and you'll spend money on the wrong layer—usually metering you can't use or automation without a proven load to control.

Scenario A: Greenfield or major expansion — start with the metering architecture

If you're building new, the metering layer is the cheapest thing you'll ever install. You're paying for the incremental cost of meters, current transformers, and comms cable while the walls are open. Retrofit that same system in three years and you're paying electricians to pull cable through finished ceilings and fire-rated walls.

But "install meters" isn't the advice. The advice is to define what the data is for before you pick hardware. In my experience, that step gets skipped about half the time, and it shows up later as either over-specified meters or a data stream nobody uses.

Match accuracy class to use case, not to habit

Revenue-grade meters (Class 0.2S or 0.5S under IEC 62053-22) belong where money changes hands—tenant billing, internal cost allocation with real budget consequences. Diagnostic-grade (Class 1) is fine for most equipment-level points: compressed air branches, chillers, individual production lines.

We rejected a submetering package last year because the spec called for revenue-grade meters on every compressed-air branch. The customer was doing internal cost allocation, not tenant billing. The accuracy class was two levels above what the use case needed, and the meter budget reflected it. Nobody could explain why.

Pick the protocol before the vendor

Modbus TCP is the lingua franca of industrial energy devices. BACnet/IP if you're tying into a building automation system. MQTT if the data is headed to a cloud platform. OPC UA if you're integrating with a modern SCADA stack.

The question to ask is not "what do you support" but "what do you support natively, without a gateway." Gateways are fine. They're also one more thing to fail, patch, and explain to IT.

We build one at GridSense—a smart submetering solution for factories—so I'll flag the bias where it comes up. But the architecture questions apply to any vendor. If the answer to "how does data leave the device" is a CSV export, that tells you something about the platform's maturity.

Scenario B: Existing plant, painful demand charges — start with demand response

Pull out twelve months of utility bills. If demand charges are a meaningful slice—enough that a 10% reduction would show up as a real line item—then demand response is the first move, not metering.

Why? Because the savings are visible fast, and they fund the rest of the program. A metering project takes a year to design, install, and integrate. A demand response program can show up on the next bill cycle.

Prove the shed capability manually before you automate

This is the part that gets skipped. Vendors love to sell the automated platform first. Automation is good. Automation without a proven shed capability is just a dashboard with a countdown timer.

The sequence I'd recommend:

1. Build a load inventory. What can actually be shed? Non-critical HVAC, EV charging, batch processes that can float 30–60 minutes, air compressors if receiver storage is sized correctly.
2. Run manual demand response for one quarter. Someone watches the demand meter and turns things off. Document what actually gets shed, for how long, and what operations pushes back on.
3. Then automate. Now the automation has a target to hit, not a target to estimate.

I've seen factories buy a demand response system and discover their real shed capability was less than half the vendor's estimate. The estimate was based on nameplate data, not on what operations would actually allow at 2pm on a Tuesday.

GridSense builds demand response controls, so apply the usual bias correction here. The sequencing advice holds regardless of platform: manual first, automated second. The vendor estimate is a hypothesis. Your first quarter of manual shedding is the test.

Scenario C: Power factor penalties on the bill — start with PFC, but verify harmonics first

If the utility is charging for power factor or billing on kVA demand, this is often the fastest payback of the three scenarios. The penalty is in dollars. You can model the correction.

Which is exactly why people rush it—and why I've rejected more PFC proposals than any other category.

Don't order capacitor banks on day one

Industrial loads today are full of harmonics: variable frequency drives, UPS systems, LED drivers, rectifiers. Harmonic currents heat capacitor dielectric and can fail banks early. This isn't a rare edge case anymore. It's the default condition in most plants with modern motor controls.

IEEE 519-2014 defines harmonic distortion limits at the point of common coupling, with the intent of limiting the distortion one customer imposes on the shared supply. The standard exists because harmonics propagate—they don't stay in your plant.

The practical sequence:

1. Meter the load with a power quality analyzer that logs individual harmonics, not just aggregate power factor.
2. If distortion is low, conventional capacitor banks are fine.
3. If harmonics are significant, specify detuned reactors or an active filter.
4. Re-measure after commissioning. Verify the fix, don't assume it.

A proposal we rejected: supplier quoted a standard capacitor bank for a plant with six large VFDs on the same bus. No harmonic study. No detuned reactors. The bank would have run hot from day one, and the failure would have been blamed on the capacitors rather than the spec.

This is also where the smart energy meter choice matters. A meter that only reports power factor and kWh won't tell you why the power factor is bad. You want one that logs individual harmonics (3rd, 5th, 7th, 11th, 13th) and can capture waveform data during switching events. The diagnostic capability is worth more than the accuracy class in this scenario.

How to figure out which scenario you're in

Three questions, one utility bill. Ten minutes.

1. Are you building or retrofitting? New construction → metering architecture first. Existing plant → keep reading.
2. What share of the bill is demand charges? If a 10% reduction would show up as a real line item, demand response is the first move.
3. Is there a power factor penalty or kVA-based charge? Yes → PFC first, with the harmonic check before you buy anything.

If two apply—common in larger plants—sequence them. Metering first if you need data to size the other two. Otherwise, start with whichever has the clearer dollar figure on the bill. That's usually PFC.

Looking back, I should have asked for twelve months of utility bills before the first technical meeting on every one of those 200 proposals. Half of them were solving a problem the plant didn't have.

Once you know your scenario, here's how to engage a manufacturer

A useful mental model for how an industrial energy management system manufacturer actually works: most aren't starting from a blank sheet. They're configuring a platform—meter families, comms modules, software layers. That's not a criticism. It's how the economics work.

But it means you should arrive with a spec, not a question. If you ask "what do you recommend," you'll get the configuration that's easiest to quote, not the one that fits your scenario.

Questions worth asking:

• What protocols does the gateway speak natively, without additional hardware?
• Can the meter log harmonics and waveform captures locally, or does that require cloud connectivity?
• What's the firmware update path, and does it require downtime?
• How does data get into our existing SCADA or EMS—push, pull, or something custom?
• What accuracy class, and is it certified to IEC 62053-22 or an equivalent?

If a manufacturer can't answer those in writing, that's information too.

The part where price stops being the point

In my experience managing reviews across a few hundred projects, the lowest quote has cost us more in roughly 60% of cases. Not because cheap equipment is bad. Because cheap equipment is often scoped for the wrong scenario, and the correction costs more than the difference.

The cheapest submetering package I've seen quoted used meters without harmonic logging. Six months later the plant needed PFC diagnostics and had to add a separate power quality analyzer. The savings on the meter were gone, plus the cost of a second installation.

That's the pattern. Not "cheap is bad." Cheap applied to the wrong layer is expensive.

So start with the bill. Pick the scenario. Then spend.