AI use case

AI analytics for smart meter and AMI data

AI that turns the flood of readings from smart electricity, gas and water meters into usable information: it monitors meter and network health at scale, estimates which appliances drive a household's usage from the meter signal alone, flags unusual consumption, and targets efficiency and electrification programmes at the customers who will benefit most, instead of a utility treating every meter and every customer the same way.

By Len Debets · Last verified 28 September 2026 · 2 public deployments

USD 600,000 to USD 10 million
Indicative value per year
A utility with 2 million smart meters and an energy efficiency programme budget. Worked example, see how it is calculated.

What problem does it solve?

Con Edison's rollout of 5.3 million smart meters was expected to generate between 100 terabytes and 1 petabyte of data annually, far more than any team can review reading by reading. Buried in that volume are the installation and configuration issues that leave a meter or its network module unhealthy, which is what Con Edison's deployment set out to find. More generally, the same kind of data holds the individual appliances, such as electric vehicles, heat pumps and air conditioning, that a utility needs to understand as adoption grows, and the customers who would benefit most from an efficiency programme but are hard to identify from billing data alone.

How does it work?

  1. Ingest at scale. Readings stream from millions of meters into a data platform, alongside the utility's asset, billing and programme data.
  2. Watch meter and network health. Machine learning flags meters and communication modules showing signs of a deployment or configuration problem, so field crews fix the ones that actually need attention.
  3. Disaggregate usage. A separate model estimates which appliances, such as HVAC, water heating, electric vehicle charging or pool pumps, are driving each home's consumption from the meter signature alone, without a sensor on the appliance itself.
  4. Segment and target. The disaggregated data groups customers by what is actually happening in their home, such as households with an electric vehicle or an ageing HVAC system, so an efficiency, demand response or electrification programme can be targeted instead of broadcast to everyone.
  5. Feed operations and customer teams. A prioritised list, a dashboard or a personalised message reaches the team or the customer, closing the loop from raw meter data to action.
Audience
Back office
Autonomy
Assist
Adoption
Early adopters
Channels
Internal tools, API and system to system

What is it worth?

Benchmarks are computed from the public deployments below: one data point per organization per KPI, with who made each claim.

No public deployment has disclosed a measurable outcome yet.

Value drivers: Lower cost to serve, Customer experience, Employee productivity.

Indicative value

A utility with 2 million smart meters and an energy efficiency programme budget

USD 600,000 to USD 10 million

Avoided field visit and rework cost from early meter health detection per year

How this is calculated

Formula: meters * issueRate * costPerIssue. The low scenario uses every low input, the high scenario every high input.

InputLowHighBasis
Smart meters in the analytics programme meters, smart meters1,500,0002,500,000Range set around the 2 million meter reference organization, well within Con Edison's 5.3 million meter deployment.
Share of meters with a deployment or health issue found by analytics each year issueRate, fraction of meters0.0050.02Editorial assumption, replace with your own meter health data.
Cost saved per flagged issue by folding it into a planned visit instead of a repeat or emergency visit costPerIssue, USD per issue80200Editorial assumption for the difference between a planned and an emergency or repeat US utility field visit; replace with your own figure. Neither deployment on this page reports a per issue avoided cost.

What it leaves out: Meter health savings only, and it assumes a flagged issue can be folded into a planned visit rather than always triggering a separate site visit, which is an editorial assumption neither deployment on this page confirms. It also leaves out the analytics platform and integration cost, and any separate value from the efficiency and electrification programmes the same data supports, which neither deployment on this page reports as a single company wide figure.

Who already uses it?

2 public deployments, strongest evidence first. Grades: A regulator or audit, B the organization itself, C vendor case study, D anonymous or estimate.

Southern California Gas Company (SoCalGas)

United States · Energy and utilities · 2020

ProductionGrade C

Southern California Gas Company (SoCalGas) worked with Bidgely on a digital only home energy report programme for medium consumption residential gas customers, a segment that traditional paper based home energy reports, which target high consumption customers, do not reach. The reports were built on AMI meter disaggregation, and the programme exceeded its savings goal, saving 565,000 therms by December 2020, with more than 405,000 customers receiving the reports digitally at a 50 percent open rate.

No outcome disclosed.

Consolidated Edison (Con Edison)

United States · Energy and utilities · 2019

ProductionGrade C

Con Edison built an enterprise data analytics platform on C3 AI to run Advanced Metering Infrastructure operations for its 5.3 million meter smart meter rollout, aggregating two years of data from 13 source systems covering 5 million customer accounts and integrating over 180 billion rows of data a year across those systems. Two machine learning algorithms and 50 analytics identify deployment and installation issues and determine meter and network health, giving the utility a real time, prioritised view from an individual meter up to the whole system, in a 10 month project.

No outcome disclosed.

How do you implement it?

A model agnostic playbook: what to prepare, the order to build in, and what goes wrong.

Data you need

  • Interval meter reads at the frequency the advanced metering infrastructure supports
  • Meter and network asset data, such as install date, model and communication module
  • Customer and premise data linking a meter to a household or business

Systems to integrate

  • Meter data management system
  • Geospatial information system for network topology
  • Customer information system or CRM, for targeting and outreach
  • Demand side management or efficiency programme platforms

Complexity: High

Con Edison's deployment aggregated two years of data from 13 source systems covering 5 million customer accounts into an integrated data image; that integration work is typically harder than the machine learning itself.

  1. 1

    Start with meter and network health, not customer analytics

    Con Edison's first phase covered deployment and installation issues plus meter and network health, using two machine learning algorithms and 50 analytics; the company planned further customer insight and distribution and transmission automation applications for later phases.

  2. 2

    Unify the source systems before modelling

    Con Edison's deployment integrated two years of data across 13 source systems before the machine learning models were configured; budget and staff for that integration effort, not just the model.

  3. 3

    Disaggregate before you personalise

    Appliance level disaggregation from the meter signal is what lets a programme target the households with an electric vehicle or an ageing HVAC system, rather than mailing everyone the same offer.

  4. 4

    Close the loop into an actual programme

    Southern California Gas Company's deployment fed a digital only home energy report programme for its medium consumption gas customers, which exceeded its savings goal, not a dashboard; without a programme on the receiving end, better analytics changes nothing for the customer.

  5. 5

    Scale meter by meter, programme by programme

    Widen coverage and add new applications, such as electrification planning or peak forecast, once the foundational data platform is proven.

Guardrails

  • Customer usage data used for targeting is handled under the utility's own data privacy and retention rules
  • A person reviews the priority list before a field crew is dispatched based on a flagged meter
  • Model changes are tested against historical data before being applied to live operations

KPIs to instrument

  • Meters flagged with a health or deployment issue, and the share confirmed correct on inspection
  • Programme enrollment and savings among AI targeted customers versus a general mailing
  • Time from a flagged issue to resolution

Human in the loop

Operations analysts triage the prioritised list of flagged meters before a field crew is dispatched, and programme managers decide which segments an efficiency or electrification campaign actually targets based on the disaggregated data.

Common failure modes

A flood of technically correct but unprioritised flags
Millions of meters can produce more anomalies than any team can act on; rank by impact and route only the highest priority batches to a person. Con Edison's application produces a prioritised list of meters that require attention.
Disaggregation that does not hold up across meter and appliance types
Appliance signatures vary by region, climate and equipment age; validate disaggregation accuracy on a local sample before using it to target a programme at scale.

What are the risks and rules?

EU AI Act

Depends on design

Annex III point 2 covers AI systems intended to be used as a safety component in the management and operation of critical digital infrastructure and the supply of water, gas, heating or electricity. Meter health prioritisation and usage disaggregation for programme targeting are not intended as safety components, so they stay outside that scope regardless of whether a person reviews the output. The tier would instead be high risk if the same kind of analytics were intended as a safety component in network operation or supply, for example directly controlling grid or metering protection systems; a human in the loop is then an Article 14 obligation for that high risk system, not a way to fall outside the category.

Controls to put in place

  • A documented separation between advisory analytics and any system that can act on grid protection or metering infrastructure directly
  • Data minimisation and retention limits on the household level usage data used for disaggregation and targeting

Frequently asked questions

What does AI actually do with smart meter data?
The two deployments on this page show different jobs on the same underlying data. Con Edison used C3 AI to monitor the health of its 5.3 million meter rollout end to end, from an individual meter up to the whole system, while Southern California Gas Company used Bidgely's platform to disaggregate usage from smart meter data for digital energy efficiency reports for its medium consumption gas customers, a programme that exceeded its savings goal.
How much data is involved?
Con Edison's smart meter deployment was expected to generate between 100 terabytes and 1 petabyte of data a year. Separately, and only once, building the analytics platform itself meant aggregating two years of data from 13 source systems covering 5 million customer accounts.
Does this replace smart meters themselves?
No. It is the analytics layer on top of an existing advanced metering infrastructure rollout; both deployments on this page assume the meters are already reporting interval data.
Can it detect theft or non technical losses?
Detecting unusual consumption patterns is a well documented research area for smart meter analytics, but neither deployment on this page reports a theft or loss detection result with a checked figure, so treat that specific claim as unproven until you have your own evidence.

How to cite this page

Blits.ai AI Use Case Library, "AI analytics for smart meter and AMI data", last verified 28 September 2026, https://www.blits.ai/ai-use-cases/smart-meter-analytics. Licensed under CC BY 4.0. Method: how we verify use cases.

Changelog
  • 28 September 2026: First published

Related use cases

Healthcare

AI command center for hospital bed and staff capacity planning

An AI powered operations center that predicts patient admissions, discharges and transfers across a hospital or health system, and helps a team of coordinators sitting in one room sequence real time bed assignments, staffing levels and patient moves, so patients get into the right bed faster and existing capacity is used fully without adding beds.

Deployments
2 public, best grade B
Reported cycle time reduction
38%
Johns Hopkins Medicine, organization claim
Retail and ecommerce

AI demand forecasting and automated replenishment for retail

Machine learning that forecasts demand for every item in every store or fulfillment center, day by day, from sales history, promotions, prices, weather and local events, and turns the forecast into automatic store and warehouse orders within limits set by planners, who handle the exceptions.

Deployments
4 public, best grade B
Autonomy
Supervised agent
Wealth and asset managementBanking

AI portfolio drift monitoring and rebalancing proposals

Continuous monitoring of every client portfolio against its mandate or model, which detects drift beyond agreed bands and prepares a tax aware, low turnover rebalancing proposal with its rationale for an advisor or portfolio manager to approve before any trade is placed.

Deployments
3 public, best grade B
Autonomy
Copilot
Pharma and life sciences

AI native platform for drug target discovery and molecule design

An AI native research platform that prioritizes disease targets from biological data, generates and optimizes candidate drug molecules computationally, and predicts their properties before a chemist synthesizes and tests them, so a pharmaceutical or biotech company reaches a validated preclinical candidate with far fewer molecules made and tested than a conventional medicinal chemistry program.

Deployments
2 public, best grade B
Reported cycle time reduction
about 60%
Insilico Medicine, organization claim