# Agent context: Electricity Theft and Non-Technical Losses

Source: https://aiplaybooks.app/industries/utilities/revenue-protection/electricity-theft
AI suitability: conditional

## Objective

Reduce non-technical electricity losses by metering the largest customers, locating losses through feeder and transformer energy balance, and ranking premises for field inspection, so inspectors find more theft, faulty meters, and billing errors per visit without accusing customers on data alone.

## Constraints

- Never bill, disconnect, or accuse a customer on the basis of data, an alarm, or a model score. A field inspector confirms every irregularity on site first.
- Record theft, faulty meters, and billing errors as separate findings, and do not train one label on all three.
- Record every inspection, including those that found nothing, with the reason the premises was chosen.
- Keep a share of each inspection campaign random, and compare every targeting method's hit rate with it.
- Use only features recorded before the inspection date.
- Use smart meter data only at the granularity and for the purposes the market's privacy rules allow for that customer type.
- Check customer-to-transformer connectivity before using an energy balance to direct inspections.
- Follow the regulator's requirements on vulnerable customers and safety before any disconnection, and report dangerous installations to the network operator.

## Inputs

- Meter readings and events per premises, at the granularity the utility may lawfully use.
- Energy measured at feeders and distribution transformers, and the customer connectivity model.
- Customer class, activity, contracted capacity, tariff, and location.
- Inspection history with reason chosen and finding, plus random inspection results.
- Meter installation records including multipliers, and occupancy and generation records.

## Outputs

- Daily exceptions for large customers, each naming the reading or event that triggered it.
- Non-technical loss per feeder and transformer, ranked, with the estimate of technical losses used.
- A ranked inspection list per campaign, sized to inspector capacity, with the main reasons for each premises and the random share marked separately for analysis but not for the inspector.
- A campaign report of hit rate and recovered energy by method, split into theft, faulty meters, and billing errors.

## Success criteria

- Non-technical losses fall in the network areas targeted, measured by energy balance.
- The targeted hit rate beats the random inspection hit rate in the same campaigns.
- Hit rates hold in neighbourhoods and customer classes that were rarely inspected before.
- Disputed findings that are overturned and safety incidents from irregular connections do not rise.

## Known pitfalls

- Training a model only on premises that were inspected, so it learns where inspectors used to go rather than where theft happens.
- Treating every non-technical loss as theft, when faulty meters, wrong multipliers, and billing errors need a meter exchange or a billing correction rather than an accusation.
- Judging a model by accuracy on a public benchmark or on simulated theft, which says little about its hit rate on real inspections.
- Running an energy balance on wrong customer-to-transformer mapping, which sends inspectors to the wrong premises.
- Acting on smart meter tamper alarms alone, which a revenue protection vendor said could be as much as 95% false in its experience.
- Disconnecting a vulnerable customer or one whose installation is unsafe without the protections the regulator requires.

## When not to use AI here

- The largest customers are not yet on tamper-resistant meters, which protects the most revenue for the least effort.
- There are too few inspection results to learn from, or they all come from campaigns targeted at the same areas and customer classes.
- Losses are concentrated in areas an energy balance already identifies, where a sweep of every premises in the area is simpler than a ranked list.
- Losses come from collection failures, collusion, or weak enforcement, which no model of consumption will change.
- The utility cannot use the granularity of meter data a model needs, under the privacy rules that apply to it.

## Candidate approaches

- Tamper-resistant metering for the largest customers (primary): Fitting the small group of industrial and commercial customers that produce most of a utility's revenue with remotely read, tamper-resistant meters and checking their readings and events every day, before spending effort on the mass of small customers.
- Feeder and transformer energy balance (primary): Comparing the energy measured entering each feeder or distribution transformer with the sum of the customer meters it supplies, so revenue protection teams work in the network areas where unbilled energy is concentrated.
- Inspection targeting model (supporting): A model trained on the utility's own field inspection results that ranks premises by their likelihood of an irregularity, so each inspection campaign visits the premises most likely to have theft or a faulty meter, with a share of random visits to correct for where inspectors used to go.
- Consumption anomaly screening (discovery): Unsupervised screening of smart meter consumption for premises whose pattern has changed or differs from similar customers, used where inspection history is thin to find candidates for inspection, many of which turn out to be faulty meters rather than theft.

## Human oversight

A field inspector confirms every irregularity on site and records the evidence, and nothing is presumed from data alone. Revenue protection staff decide back-billing, disconnection, and any referral to the police, taking into account the customer's vulnerability and making the installation safe. A model or alarm only ranks premises for inspection, and a revenue protection analyst reviews why a premises was ranked before it goes into a campaign.

## Sources

- [Can Utilities Realize the Benefits of Advanced Metering Infrastructure? Lessons from the World Bank's Portfolio](https://documents1.worldbank.org/curated/en/979981468940285618/pdf/107120-BRI-ADD-SERIES-Live-Wire-Series-09-Knowledge-Note-PUBLIC.pdf), World Bank, Live Wire knowledge note, 2016, practitioner
- [Deputizing Data: Using AMI for Revenue Protection](https://www.renewableenergyworld.com/hydro-power/dams-civil-structures/deputizing-data-using-ami-for-revenue-protection/), Renewable Energy World, quoting a revenue protection analytics vendor, 2008, case-study
- [The Challenge of Non-Technical Loss Detection Using Artificial Intelligence: A Survey](https://doi.org/10.2991/ijcis.2017.10.1.51), International Journal of Computational Intelligence Systems, 2017, research
- [Is big data sufficient for a reliable detection of non-technical losses?](https://doi.org/10.1109/isap.2017.8071420), International Conference on Intelligent System Application to Power Systems, 2017, research
- [Detection of Non-Technical Losses Using Smart Meter Data and Supervised Learning](https://doi.org/10.1109/tsg.2018.2807925), IEEE Transactions on Smart Grid, 2018, research
- [A case study of improving a non-technical losses detection system through explainability](https://doi.org/10.1007/s10618-023-00927-7), Data Mining and Knowledge Discovery, 2023, research
- [A Human-in-the-Loop Approach based on Explainability to Improve NTL Detection](https://doi.org/10.1109/icdmw53433.2021.00123), IEEE International Conference on Data Mining Workshops, 2021, research
- [Electricity Theft Detection in AMI Using Customers' Consumption Patterns](https://doi.org/10.1109/tsg.2015.2425222), IEEE Transactions on Smart Grid, 2015, research
- [Wide and Deep Convolutional Neural Networks for Electricity-Theft Detection to Secure Smart Grids](https://doi.org/10.1109/tii.2017.2785963), IEEE Transactions on Industrial Informatics, 2018, research
- [Power distribution utilities record positive PAT after years of losses](https://www.pib.gov.in/PressReleasePage.aspx?PRID=2215761), Government of India, Ministry of Power, 2026, case-study
- [Tackling electricity theft, new requirements for electricity suppliers](https://www.ofgem.gov.uk/sites/default/files/docs/2014/05/tackling_electricity_theft_-_new_requirements_for_electricity_suppliers.pdf), Ofgem, 2014, regulation
- [Energy Theft Reduction Strategy](https://www.retailenergycode.co.uk/our-programmes/energy-theft-reduction-strategy/), Retail Energy Code Company, regulation
- [Access to half-hourly electricity data for settlement purposes: a Data Protection Impact Assessment](https://www.ofgem.gov.uk/sites/default/files/docs/2018/07/data_protection_impact_assessment_2.pdf), Ofgem, 2018, regulation