How Wrong Is My Smart Meter? I Measured It

My SMETS2 smart meter read 2.4% higher than a calibrated Fluke 376 FC clamp meter over 43 days of logging, with divergence concentrated in low-load periods and near the 15-minute settlement boundary—enough to cost roughly £18/year at August 2026 prices if uncorrected.

The Test Setup

I fitted the Fluke 376 FC around the meter tails on July 12, 2026, leaving my SSE-issued SMETS2 untouched. The clamp logs true-RMS current every second with ±1% accuracy plus 3 counts; I exported to CSV and resampled to match the smart meter's 30-minute intervals. My house: 3-bed semi, gas heating, EV charger on a dedicated 32A circuit. I excluded the EV circuit from the clamp reading to isolate the metered load, matching the smart meter's view.

What "Accuracy" Actually Means

Smart meters must meet MID Class B standards: ±2% for active energy between 5% and 100% of rated current, ±3% below 5%. That's a generous window. The Fluke, calibrated March 2026, carries traceability to UKAS—tighter than my meter's factory certificate, which I requested from SSE and received as a 2019 PDF with no serial-specific data. Standards compliance is not precision.

The Headline Numbers

Over 2,064 half-hour periods, the smart meter logged 284.7 kWh; the clamp recorded 278.1 kWh. The 6.6 kWh gap is 2.37% high. Daily drift varied: 0.8% on heavy-use days (washing machine, oven, power tools), 4.1% on light days when baseline sat below 100W. The smart meter's minimum detectable load appears higher than specified, or its sampling algorithm rounds upward aggressively at low draw.

Time-of-Use Penalty in the Gaps

My tariff switches rates at 00:00, 07:00, 13:00, and 19:00. I found twelve instances where the smart meter's clock drifted 2-4 minutes against my NTP-synced logger, pushing consumption into the wrong band. One 2.3kW oven cycle at 18:58 logged partly in the peak window, partly off-peak. Over six weeks, this boundary timing error cost an estimated £0.84—small, but systematic. Shift workers and night-rate EV chargers would see worse.

Load-Type Patterns

Resistive loads (kettle, immersion) showed 1.1% divergence, within spec. Inductive loads—fridge compressor, washing machine motor—spiked to 3.8% during start-up transients. The smart meter's 1-second sampling misses brief 800W surges that the clamp catches. LED lighting with poor power factor (measured 0.52 on a cheap B22 bulb) pushed divergence to 5.2%. Power factor correction is not mandatory for domestic meters; the error accumulates.

The Privacy Trade-Off Nobody Asked For

My smart meter transmits 30-minute profiles to DCC, then to SSE. I requested my data under GDPR; it arrived as 16,384 rows of CSV with 14-day latency. The clamp data stays local—Bluetooth to phone, no cloud. This matters because granular load data reveals occupancy patterns, appliance signatures, even which TV channel you're watching based on screen brightness draw. The accuracy question is inseparable from who holds the accurate data.

What I Did With the Findings

I submitted a disputed reading to SSE on August 22, 2026, citing my calibrated log. They requested the Fluke's calibration certificate (provided) then offered a 1.2% bill credit as "goodwill," not admission of error. I declined. Instead, I shifted my tariff to a simpler single-rate fixed deal where boundary timing errors cannot compound, accepting higher per-kWh cost for predictability. The £18/year potential overpayment is now moot; the time cost of disputing exceeded the recovery.

Should You Verify Yours?

Clamp meters with logging run £180-400; calibration adds £60. For typical annual consumption of 2,900 kWh, a 2% error is £35-50/year at August 2026 prices—payback in 4-7 years if you act on findings. Most won't. My recommendation: request your meter's MID certificate (free), check for firmware updates (my SMETS2 last updated October 2025), and log one heavy-use day with a borrowed clamp if your bills feel wrong. The error is real, but the cure is often bureaucratic, not technical.