What home energy monitoring actually saves

A clamp fitted round the main feed will report, correctly, that a house is drawing 2.4 kilowatts right now. It will not say whether that is the kettle, the tumble dryer or a pool pump running on a timer. That gap, between an accurate total and a useful answer, is where most home energy monitoring hardware actually lives, and where the marketing around it tends to stop short.

Three approaches are sold to close that gap, and they are not interchangeable. A pair of clamps on the incoming supply feeds a total load to software that then guesses at individual appliances. A bank of smaller clamps fitted at the consumer unit measures each circuit separately, branch by branch. A smart plug sits between one appliance and the wall socket and measures precisely that appliance, nothing else. Each gets a different question right, and each gets a different one wrong.

Three instruments, three questions

The three sit on a line running from cheap-and-vague to precise-and-narrow. Whole-panel monitoring with disaggregation needs one installation and no rewiring, but every appliance-level reading below the total is a statistical inference. Per-circuit monitoring needs a qualified installer inside the consumer unit and one clamp per circuit, but every reading is a real, measured current on a real conductor. A smart plug needs no installation beyond being plugged in, and the figure it reports is not an estimate of anything: it is what passed through that socket.

A clamp on the mains, and a guess about the rest

Sense, one of the established products in this category, clamps two current sensors round the main feed and reads voltage off the same 240V breaker that powers the unit. By the company's own description of the hardware, it samples at one megahertz, generating around four million data points a second, then looks for the switching signature of individual appliances in that stream, a technique the company says draws on methods used in speech recognition, matching patterns against a library of known devices. Nothing is measured at the appliance itself. Everything below the main total is inferred.

Sense says recognising a household's own devices can take weeks, because the software needs to see an appliance switch on and off enough times to learn its signature. A peer-reviewed review of the same broad technique, generally called non-intrusive load monitoring, backs that caution with numbers: it catalogued published mean absolute errors across research datasets running from a few watts for some appliances to several tens of watts for others, including refrigerators and dishwashers, varying by algorithm and by house. The same review noted that most published methods have not been shown to work reliably in a home outside the one they were trained on. A whole-panel monitor is honest about the total. Anything it tells you about one specific appliance is a probability, not a reading.

Per-circuit sensors trade installation work for certainty

The alternative is to put a clamp on the wire itself. Emporia's Vue, one of the more widely used per-circuit systems, uses split-core current transformers, a 200-amp clamp for the main feed and smaller 50-amp clamps for individual breakers, that open to fit around an existing conductor without disconnecting it. Every branch gets its own number, and that number is a direct current measurement rather than an inference.

Emporia's own published accuracy data is more nuanced than the headline figure usually quoted. The company states its sensors average under 1% error at unity power factor, with 99.2% of measurements falling inside a ±2% band under typical conditions. That band widens sharply as power factor drops: at moderate power factors of 0.5 to 0.7, Emporia's own documentation puts branch clamp error at roughly 4–8.5%, rising to 40% or more at the extremes. Motors, chargers and switched-mode power supplies, all common in an ordinary home, often run at exactly those lower power factors. A per-circuit reading is real. Its precision still depends on what is plugged into that circuit, not on the sensor alone.

It also only ever reports a circuit, not a device. A bedroom breaker feeding four sockets and a ceiling light will report as one figure, whatever combination of things is switched on. Splitting that figure by appliance needs either a smart plug on each item, or a slower process of elimination: switching things off one at a time and watching the number move.

A smart plug tells the truth about one thing

A smart plug avoids inference entirely. Whatever wattage it shows is what passed through that one socket, for that one appliance. The trade-off is scale. A house has dozens of sockets, and a monitor on each one is neither affordable nor tidy. Smart plugs work best aimed at a specific suspect, a router, a games console, a printer left permanently switched on, rather than deployed as a way to cover a whole property.

There is a genuine limit at the bottom of the range. IEC 62301, the international standard for measuring standby power, requires test instruments to resolve 0.01 watts for readings under 10 watts, and to hold measurement uncertainty below 2% at a 95% confidence level for anything above 0.5 watts. Consumer smart plugs are not generally built or calibrated against that standard; most resolve to a tenth of a watt at best. A phone app showing '1W' for a device left on standby is a rounded figure, not a laboratory-grade one, and the true draw could plausibly sit on either side of it.

What the sampling rate actually buys

UK government guidance on smart meters states that they record consumption data half-hourly, with the in-home display updating roughly every 30 minutes for gas and near-instantly for electricity. That is enough to see a house's daily shape: the morning peak, the evening peak, a week away with nobody home. It is not enough to see a kettle that boils and switches off inside three minutes; that event gets folded into a half-hour average alongside whatever else was running at the time. Seeing it as a distinct event needs sampling faster than the appliance's own cycle, which is why home monitors sample locally, at rates from once a second up to Sense's one megahertz, rather than relying on the utility feed at all.

Standby load, measured properly

The most rigorous study of whole-house standby power remains, oddly, a small one. In 2000, researchers at Lawrence Berkeley National Laboratory measured every standby-drawing appliance in ten Californian homes, then cross-checked the sum against a separate reading taken directly off each house's own utility meter with everything either unplugged or switched to standby. The two methods agreed closely enough that the researchers judged short utility-meter readings alone would be reliable for future studies of this kind.

Those wattages are now a quarter of a century old, and the devices behind them have mostly moved on. The televisions and video recorders that dominated the 2000 list have largely been replaced by LED sets, routers and streaming boxes with their own, usually lower, standby draws. The broad pattern the study exposed has not obviously disappeared: small idle loads, added together across a house, reaching a high single-digit share of the annual bill. But nobody should treat 67 watts, or 9%, as a figure for a modern home. It is the method that has lasted, not the number.

The honest limits

None of the three approaches changes a bill by itself. A monitor, however accurate, only reports; a household still has to notice a pattern, decide it matters and act on it. Even the accuracy of the instrument is bounded. IEC 62053-21, the standard behind the meters utilities actually bill against, defines Class 1 and Class 2 accuracy as 1% and 2% error respectively, but only at full load and unity power factor. A household clamp reading a few per cent off a utility bill is not necessarily faulty. Both instruments carry their own disclosed margin, and the two margins do not always point the same way.

A branch circuit shared by four sockets will always report as one number, however good the clamp fitted to it. Disaggregation software will always be making an educated guess below the level of the main feed, however many weeks it has had to learn a house's appliances. And nothing measures the one thing that actually produces a saving, which is a habit changed or a device replaced.

The most reliable way to find out what a single appliance costs to leave switched on is still the plainest one. Plug it into its own meter, on its own, and read the number directly off the device rather than off a model of it.