Industries / Energy, water and waste / Electric transmission and distribution / Smart metering
Data and billing
Data and billing is what happens after a reading arrives: systems check and store it, fill gaps, work out what the customer owes under the tariff, and use detailed readings to set prices that change by hour or half hour.
7
Milestones achieved, from 1992 to 2019.
2
To watch for: announced, not yet confirmed outside.
2
Open deliverables. Nobody has shipped them.
The road
AchievedWatch forOpen
Achieved
1992
Appliances read from one meter
George Hart showed in the Proceedings of the IEEE that the switching on and off of each appliance can be picked out of the single signal at a home's meter, without a sensor on every appliance. The method, now called non-intrusive load monitoring or disaggregation, came out of MIT Energy Laboratory and EPRI work from 1984 to 1985.
Credit
George W. Hart, with Ed Kern and Fred Schweppe
1993
Prepaid meters get a standard
Eskom, South Africa's state utility, began the Standard Transfer Specification (STS), under which a customer buys a 20-digit token and types it into the meter to load credit; it is now published as IEC 62055. The STS Association reports about 50 million meters run by 500 utilities in 94 countries.
Lessonthe token's time counter, set from a 1993 base date, ran out on 24 November 2024, and every STS meter had to be recoded before then or stop accepting credit.
Credit
Eskom; STS Association
2010
Bills, not meters, failed
After customers of Pacific Gas and Electric in California complained of high bills on their new smart meters, the state regulator CPUC hired the Structure Group to test them. It tested more than 750 smart meters and found them accurate, but found failures in PG&E's customer service and in bills that were cancelled and reissued several times.
Lessonaccurate meters do not win trust if billing and customer handling go wrong.
Credit
California Public Utilities Commission; The Structure Group (findings)
2012
One data store for a province
Ontario finished fitting 4.8 million smart meters by December 2012, and all of its utilities send readings to one central meter data repository run by the province's grid operator, the IESO, which returns hourly figures for billing. About 4.5 million customers were moved onto time-of-use prices.
Credit
IESO; Ontario Energy Board; Ontario government
2014
Hourly prices for households
From 1 April 2014 Spain's default regulated electricity price for homes up to 10 kW, the PVPC, follows the wholesale market and changes every hour, and households with a working smart meter are billed on their actual hourly use.
Credit
Spanish Ministry for the Ecological Transition (Royal Decree 216/2014)
2018
Half-hourly prices at home
Octopus Energy launched Agile Octopus, a household tariff with a new price for every half hour, set at 4 pm each day from wholesale prices, which pays customers when wholesale prices fall below zero. Octopus called it the world's first such tariff; it depends on a smart meter recording each half hour.
Credit
Octopus Energy
2019
Norway's central data hub
Statnett, Norway's grid operator, switched on Elhub on 18 February 2019, a single store for the readings of about 3.25 million metering points that every supplier and grid company must use.
Credit
Statnett; built by Accenture
Watch for
Britain settles every half hour
Britain's market-wide half-hourly settlement programme is moving every meter onto settlement by actual half-hourly readings, in place of estimated profiles; Ofgem has set a final migration deadline of 7 May 2027, and the programme reported 10 million meter points started, close to 30%, in May 2026.
Achieved whenElexon or Ofgem confirms the migration finished by the deadline.
Claimed by
Ofgem and the MHHS programme, April and May 2026
AI in meter data systems
Itron says IEE Cloud AI, the next version of its meter data management system, will let utilities check readings as they arrive, available worldwide from January 2027; it gives no measured results.
Achieved whena named utility publishes measured gains, such as fewer estimated bills.
Claimed by
Itron, September 2026
Open
01
A test for bad-reading checks
A public answer key for "validation, editing and estimation" (VEE), the rules every meter data system runs to catch bad readings and fill gaps: real half-hourly readings with each fault (spikes, stuck values, clock shifts, meter swaps, runs of zeros) labelled by hand, plus a scoring script.
The briefCredit · unclaimed02
Gap-filling by open AI models
A public leaderboard that tests whether time-series foundation models, AI models pretrained on many kinds of time series, fill gaps in meter readings better than the simple rules utilities use.
The briefCredit · unclaimed
Open 01 · the brief
A test for bad-reading checks
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public answer key for "validation, editing and estimation" (VEE), the rules every meter data system runs to catch bad readings and fill gaps: real half-hourly readings with each fault (spikes, stuck values, clock shifts, meter swaps, runs of zeros) labelled by hand, plus a scoring script.
Why it is still open
A search found no public VEE benchmark with labelled faults; published work on smart meter datasets covers forecasting and disaggregation. The London smart meter dataset, half-hourly readings from 5,567 households under an open license, is a ready base.
Built with what exists
The London Datastore smart meter data; Python with pandas; a labelling tool such as Label Studio; a rule-based VEE baseline written from published utility rules.
First step
Write the list of fault types and have two people label 50 meters.
Done looks like
About 500 meter-months labelled, every label agreed by two people working apart, and precision and recall published for the rule-based baseline.
Open 02 · the brief
Gap-filling by open AI models
The first to publish is credited on this line, and the deliverable moves from Open to Achieved.
What to build
A public leaderboard that tests whether time-series foundation models, AI models pretrained on many kinds of time series, fill gaps in meter readings better than the simple rules utilities use.
Why it is still open
The one benchmark found used a small sample of meters with artificial gaps, and in its own table the classical Holt-Winters method beat both Chronos and TimesFM on mean absolute error.
Built with what exists
The London Datastore data; Chronos and TimesFM, both open; Python; one GPU.
First step
Re-run the published benchmark on 10 meters with open code and compare the numbers.
Done looks like
Code and scores public for all 5,567 London households, comparing at least three models with linear interpolation and a "same time last week" rule, and a clear finding on whether any model beats the rules by at least 10% in mean absolute error.