Energy and utilities · Playbook 30

Water leak detection

Leaks found sooner, located more precisely and fixed in the right order. The models point to where the water is going; a person decides every dig, every shut-off and every call to a customer.

  • Predictive ML
  • Event-driven integration
  • Agents with tools
  • Human in the loop
  • Ofwat PR24 leakage
  • NIS Regulations 2018 (DWI)
  • Guaranteed Standards Scheme
  • Street works permits
  • UK GDPR

At a glance

The problem
Leakage is reported a year at a time, but leaks are found one at a time, often days or weeks after they start
Typical volume
Around 1,000 DMAs, 20,000 acoustic loggers, 3,000 pressure loggers and 400,000 smart meters, with 100 to 150 suspected leaks a week
Who's involved
Leakage and network operations, telemetry and OT engineers, the control room, field and street works teams, customer service, regulation, IT security, data protection
Pilot
12 weeks in 40–60 DMAs, after a 4-week discovery
Platforms
AWS, Azure, or a sovereign or on-prem stack, with telemetry and SCADA always on-prem

An illustrative scenario. The numbers are typical examples, not a client case study. The cloud services and UK rules were checked on 9 November 2026.

The business problem

A large share of the water a company puts into supply never reaches a tap. Most of it is lost through small, steady leaks on mains and customers' supply pipes, not the dramatic bursts that make the local news. The network is divided into district metered areas (DMAs), each with metered inflows. In the small hours, when almost everyone is asleep, the flow into a DMA should drop to a known minimum. When the night flow creeps up, water is going somewhere.

The signals are already there, but in different places. DMA flows and pressures come in through telemetry and SCADA. Pressure loggers and acoustic noise loggers report once or twice a day to their own platform. Smart meters report through a head-end system. GIS holds the pipes, and the work management system holds the jobs. Leakage analysts spend their mornings joining these up by hand, then send technicians out to listen for leaks in DMAs that may or may not be the right ones.

The cost is regulatory as well as physical. Leakage is a PR24 performance commitment with money attached, and companies have committed to halve leakage by 2050. Every repair also has its own consequences: a valve shut off can interrupt supply to hundreds of homes, with Guaranteed Standards Scheme payments if notice or restoration times are missed, and every dig in the road needs a street works permit. Finding leaks faster only helps if the follow-up is planned well.

What AI changes

  • Telemetry leaves the OT network one way, through a DMZ, and joins the logger and smart meter feeds in one event pipeline outside it.
  • Night-flow models compare each DMA's minimum night flow with its expected legitimate use, and flag DMAs where leakage is rising, with the size of the rise and a confidence.
  • Burst detection watches flow and pressure in near real time, and acoustic models classify logger noise as likely leak, likely not, or unclear, pointing to the stretch of main between loggers.
  • Signals about the same leak are grouped into one suspected leak and ranked by estimated volume, risk to customers and critical assets, and how sure the evidence is.
  • An agent assembles the evidence pack and drafts the work order, including the valves that might need operating and the properties and priority customers that would be affected.
  • A leakage analyst confirms every suspected leak before a job is raised. A network controller decides any valve operation or interruption, and the customer team decides who is contacted and how. The AI never operates the network.

How success is measured

Agree these, and measure the baseline, before anything is built. Example targets are typical, not promises.

Leakage against the performance commitment

Megalitres a day, on Ofwat's three-year-average measure, in the pilot DMAs and company-wide.

ExampleOn the glidepath to the 2030 commitment

Awareness time

From when a leak starts (estimated from the night-flow step) to when someone knows about it.

ExampleFrom weeks to days for leaks above an agreed size

Find rate

Share of suspected leaks where the technician finds a leak on site. Low means wasted visits and dry holes; it is the measure that builds trust.

ExampleAbove 70%, reviewed weekly

Bursts reported by customers first

Visible bursts where a customer called before the system raised it. This keeps the burst model honest.

ExampleFalling quarter on quarter

Supply minutes lost to repairs

Interruption minutes per property caused by leak repairs, split planned and unplanned.

ExampleNo worse than baseline while find rate rises

Customer-side leaks followed up

Continuous-flow alerts from smart meters that led to contact and a fix, within the company's policy.

ExampleEvery alert actioned or closed with a reason

UK regulation and controls

Cross-sector rules are on the UK regulation map.

  • OfwatPR24 performance commitments for 2025–2030: leakage and water supply interruptions

    Ofwat's PR24 package includes a 17% cut in leakage over 2025–2030, measured as a three-year average against a 2019–20 baseline, with financial incentives either way. Supply interruptions are a separate commitment, so a fix that shuts off hundreds of homes can cost more than it saves. The model helps find leaks; it never calculates the reported leakage figure, which stays with the regulatory reporting team under Ofwat's methodology and assurance. The government plans to replace Ofwat with a single new water regulator; Ofwat's rules apply until then.

  • DefraLeakage targets: halving leakage by 2050, with interim targets

    Water companies have committed to cut leakage by 50% by 2050 from a 2017–18 baseline. The government's delivery plan sets interim leakage reductions of 20% by 31 March 2027 and 30% by 31 March 2032. The design keeps the evidence for each repair, so progress can be shown leak by leak, not only in the annual total.

  • OfwatWater (Special Measures) Act 2025

    Ofwat now sets rules that can stop performance-related pay for directors where a company misses standards on customers, the environment, financial resilience or criminal liability, and obstructing a regulator's investigation is an offence. Leakage evidence gets board-level scrutiny. Keep a complete, retrievable record of every alert, decision and repair, and never let a model change a reported number.

  • Defra / DWINIS Regulations 2018 and the Cyber Security and Resilience Bill

    A company supplying drinking water to 200,000 or more people is an operator of essential services. Defra and the Welsh Ministers are the competent authorities, with the DWI acting for them. Telemetry and SCADA are part of the essential service, so the leak platform reads from them through a one-way, outbound-only path and can never write back. The Bill, still going through Parliament, will tighten incident reporting and bring managed service providers into scope.

  • DefraGuaranteed Standards Scheme (strengthened July and October 2025)

    In England, a planned interruption of more than four hours needs 48 hours' notice, or the company owes £50 to a household and £100 to a business. If supply isn't restored within 12 hours of an unplanned interruption, a household is owed £50 and a further £50 for each 12 hours after. A separate standard covers support for customers on the priority services register. The work order shows the properties and priority customers a shut-off would affect, and the notice clock, so a person can plan it.

  • DfT / highway authoritiesNew Roads and Street Works Act 1991 and permit schemes

    Every excavation needs a permit or notice. Urgent leak repairs are immediate works with a short notification window, and since 5 January 2026 fixed penalties have doubled and overrun charges apply at weekends and bank holidays. The AI never books or starts works. Leaks that can wait are grouped so one planned permit covers several repairs.

  • ICOUK GDPR and the Data Protection Act 2018: smart meter data

    Household consumption at hourly or finer intervals is personal data, and can show when people are home. Use it for leak detection under a stated lawful basis and privacy information, aggregate it to DMA level wherever that is enough, and keep it only as long as needed. A DPIA comes before the pilot. Contacting a customer about a leak on their side is a person's decision, following the company's published policy.

The solution

Showing AWSAzuresovereign / on-prem services. Switch with the toggle; every diagram and table on this page follows it.

Water leak detection: reference architecture
  1. 01Network and telemetry (OT)

    • DMA flow meters, pressure sensors and outstations
    • SCADA and telemetry historian
    • Pressure and acoustic loggers (logger platform)
  2. 02Edge and intake

    • AWS IoT GreengrassAzure IoT Operations on Arc-enabled KubernetesOPC UA client and MQTT broker (e.g. Eclipse Mosquitto) in the DMZOne-way export through the DMZ
    • API Gateway + Kinesis Data StreamsAPI Management + Event HubsKong + Apache KafkaLogger and smart meter feeds
    • Timestream for InfluxDBFabric Real-Time Intelligence (Eventhouse)TimescaleDB or InfluxDBTime-series history
  3. 03Intelligence

    • SageMaker AIAzure Machine LearningMLflow and Kubeflow on KubernetesNight-flow and burst models
    • SageMaker AIAzure Machine LearningMLflow and Kubeflow on KubernetesAcoustic leak classifier
    • Agent
      Bedrock AgentCoreFoundry Agent ServiceLangGraph on Kubernetes, with open-weight models via vLLMLeak investigation agent
  4. 04Orchestration

    • Step FunctionsLogic Apps / Durable FunctionsTemporal, or n8nSuspected-leak workflow
    • EventBridge + SQSEvent Grid + Service BusApache Kafka or RabbitMQEvents and queues
  5. 05Systems of record

    • GIS and hydraulic model
    • Work management system
    • Customer and billing system, with priority services register
  6. 06People

    • Person
      Leakage analyst: confirms and prioritises
    • Person
      Network controller: decides valves and interruptions
    • Person
      Customer and street works teams
  • Cloud or platform service (AWSAzuresovereign / on-prem)
  • AI agent
  • A person decides
  • Your existing systems
  • Outside your organisation

Components, side by side

Water leak detection: components on AWS, Azure and a sovereign or on-prem stack
LayerAWSAzureSovereign / on-prem
Reading telemetry out of OTRead from a historian replica in the DMZ, not from outstations. Read-only accounts and nothing that can write back. If a site still runs Azure IoT Edge, 1.5 LTS support ends on 10 November 2026; 1.6 LTS runs to November 2028.AWS IoT Greengrass in the DMZ, reading a historian replica or OPC UAAzure IoT Operations (OPC UA connector, MQTT broker, data flows) on Arc-enabled KubernetesAn OPC UA client and an MQTT broker (e.g. Eclipse Mosquitto) on Kubernetes in the DMZ
Logger and smart meter intakeLoggers and meter head-ends push batches once or twice a day. Expect late and duplicate data, and design for it.API Gateway, Kinesis Data StreamsAPI Management, Event HubsKong or your existing API gateway; Apache Kafka
Time-series historyTimestream for LiveAnalytics is closed to new customers. AWS IoT SiteWise isn't offered in London, and IoT Events has ended, so neither is used.Timestream for InfluxDBFabric Real-Time Intelligence (Eventhouse)TimescaleDB or InfluxDB
Night-flow, burst and acoustic modelsStart with per-DMA night-flow baselines and simple burst rules, then add acoustic classification. Labels come from repair records, which need cleaning.Amazon SageMaker AIAzure Machine LearningMLflow and Kubeflow on Kubernetes
Asset and network contextPipe material, age, burst history and valve locations stay in GIS. The models read them; they don't keep their own copy of the network.Your GIS and hydraulic model, through their APIsYour GIS and hydraulic model, through their APIsYour GIS and hydraulic model, through their APIs
Procedures and past repairs (retrieval)Network operations procedures, valve operation rules and past job notes, so drafts cite the procedure rather than guess.Bedrock Knowledge BasesFoundry IQ, built on Azure AI SearchOpenSearch, or PostgreSQL with pgvector
Reasoning and draftingBoth clouds offer a choice of models, including Claude. The language model drafts; the detection comes from the ML models.Amazon BedrockMicrosoft FoundryOpen-weight models served with vLLM on your own GPUs, or IONOS AI Model Hub
Agent runtimeThe original Bedrock Agents closed to new customers on 30 July 2026; new builds use AgentCore.Amazon Bedrock AgentCoreFoundry Agent ServiceAn open-source agent framework (e.g. LangGraph) on Kubernetes
Workflow and eventsAWS Step Functions, EventBridge, SQSLogic Apps or Durable Functions, Event Grid, Service BusTemporal or n8n; Apache Kafka or RabbitMQ
Link to work management and customer systemsDrafts land in a draft status the agent can't change. Reuse the integration layer you have.API Gateway, plus your existing integration platformAPI Management, plus your existing integration platformKong or your existing API gateway, plus your existing integration platform
Human reviewAnalysts confirm in the screens they already use. No new console for the control room.In the leakage team's existing tools and the work management systemIn the leakage team's existing tools and the work management systemIn the leakage team's existing tools and the work management system
Identity, keys, OT securitySeparate identities for OT and IT, and no shared accounts across the boundary.IAM Identity Center, KMS, device certificates in AWS IoT CoreEntra ID, Key Vault, Azure Device Registry certificatesKeycloak or your existing Active Directory; OpenBao or HashiCorp Vault; an HSM
Monitoring and auditLog every alert, every analyst decision and every repair outcome, with the model version used.CloudWatch, CloudTrailAzure Monitor, Log AnalyticsPrometheus, Grafana, OpenTelemetry
UK data residencySmart meter data is personal data. Azure IoT Operations connects to Azure Arc through regions outside the UK; check what metadata goes where.eu-west-2 (London)UK South / UK WestYour own UK data centre, or a UK or EU provider

On a phone, the table shows the platform you pick. Sovereign / on-prem means open, portable components you can run in your own data centre or with a UK or European provider; IONOS is shown as one example, and others exist. What sovereign cloud means →

Hybrid and on-prem

Many organisations keep core systems and sensitive data on their own servers. This is how the same design works across both.

Stays on-prem

SCADA, telemetry outstations, the control room systems and the historian stay on the OT network, untouched. So do GIS, the hydraulic model and work management if they run on the company's own servers today.

Uses cloud AI

The time-series history, model training, the night-flow and acoustic models and the investigation agent run in a UK cloud region, or on the company's own servers with open-weight models. Only flows, pressures, logger results and meter readings leave OT, never control logic.

How they connect

A gateway in the DMZ sends telemetry out over an outbound-only, encrypted connection, over AWS Direct Connect, Azure ExpressRoute or the internet with mutual TLS. Logger and meter platforms push to the API layer. Nothing outside can open a connection into OT.

If the link drops

The network and the control room don't notice. The DMZ gateway buffers data and catches up later. Analysts fall back to their existing night-flow reports and logger platform, and burst alarms still reach the control room through SCADA, as they do today.

The agents

What each agent may do through its tools, what it may not, and where a person signs off. The limits are enforced by the platform, not the prompt.

Leak investigation agent

May do

  • Read the suspected leak, the model scores and the DMA's flow, pressure and logger history
  • Look up the pipes, valves, burst history and critical customers in GIS
  • Count the properties and priority-services customers a valve shut-off would affect
  • Find related open jobs, recent repairs and street works already planned nearby
  • Draft a work order in draft status, with the evidence and a suggested priority

May not do

  • Send anything to SCADA, telemetry or any system on the OT network
  • Release, approve or close a work order
  • Plan or request a valve operation or supply interruption
  • Apply for a street works permit or book a crew
  • Contact a customer or change a customer record

Where a person signs offA leakage analyst confirms every suspected leak. A network controller decides any valve operation or interruption, and the customer team decides customer contact. The agent's permissions are enforced by the platform's identity and policy controls and by the network design, not by instructions in a prompt.

Middleware and integration

Data flows one way out of OT. The DMZ gateway reads a historian replica with read-only accounts and publishes over MQTT. There is no return path, which is enforced by firewall rules, and a data diode is worth considering for the busiest sites.

The event pipeline joins four kinds of signal: a DMA night-flow rise (a daily batch once the 2am to 4am data is in), a burst transient on flow or pressure (within minutes), an acoustic logger hit (when the logger uploads), and a continuous-flow alert from a smart meter. Each event carries an ID, location and time window. A grouping step merges events about the same stretch of main into one suspected leak, so a retry or a late logger upload never creates a second job.

Work management and customer systems are reached only through their APIs and the existing integration layer. If work management is down, drafts wait safely in a queue. The AI can be switched off without touching the network: leakage work carries on from the existing reports, as it did before.

Process and people

  • A leakage manager owns the models and thresholds for each DMA type, with change control like any other operational system.
  • A daily review of new suspected leaks during the pilot, with technicians' findings fed back as labels.
  • Override rules: analysts record why they rejected or reprioritised a suspected leak, and the reasons are reviewed monthly.
  • The OT security owner signs off every new connection, and the isolation plan is tested before go-live.
  • Valve operations and interruptions follow the existing network operations procedures and control room authority, with Guaranteed Standards and priority customers checked before any planned shut-off.
  • Regulatory reporting stays separate: the reported leakage figure is calculated under Ofwat's methodology by the reporting team, and assured as now.

Delivery plan

  1. 4 weeks

    Discovery

    Map today's leakage process, check telemetry, logger and meter data quality in candidate DMAs, clean a year of repair records, and agree the OT/IT boundary.

  2. Gate · must be true to passPilot DMAs chosen, data access agreed with telemetry engineers, OT security review and DPIA started, baseline find rate and awareness time measured.

  3. 12 weeks

    Pilot

    40–60 DMAs, in shadow mode: suspected leaks go to the leakage team only, and are compared with what technicians find.

  4. Gate · must be true to passFind rate and awareness time within agreed limits, no route into OT found in testing, and the leakage manager signs off.

  5. 4–6 weeks

    Production

    Draft work orders go live in the pilot DMAs, with monitoring, support and a rollback plan.

  6. Gate · must be true to passIsolation plan tested, support model in place, analysts and controllers trained.

  7. Ongoing

    Scale

    Add DMAs in waves, then customer-side leak alerts as smart meter coverage grows.

  8. Gate · must be true to passEach wave passes the same pilot gate; each new data source passes its own security and data protection review.

Roles and RACI

Who is responsible, accountable, consulted and informed, by activity
ActivityDirector of water networksLeakage managerOT / telemetry engineerSolutions architectI can fill thisTechnical PMI can fill thisEngineersI can fill thisSecurity (NIS)I can fill thisDPO
Business case and benefitsAccountableResponsibleConsultedConsultedConsultedInformedInformedInformed
Target architecture and OT/IT boundaryInformedConsultedConsultedAccountable and ResponsibleConsultedConsultedConsultedInformed
NIS and OT security reviewInformedConsultedResponsibleConsultedInformedConsultedAccountableInformed
Telemetry, logger and meter data accessInformedConsultedAccountable and ResponsibleConsultedConsultedResponsibleConsultedConsulted
DPIA for smart meter dataInformedConsultedInformedConsultedInformedInformedConsultedAccountable and Responsible
Build and integration with GIS and work managementInformedConsultedConsultedAccountableResponsibleResponsibleConsultedInformed
Model testing against repair historyInformedAccountableConsultedConsultedConsultedResponsibleInformedInformed
Go-live decisionAccountableResponsibleConsultedConsultedResponsibleInformedConsultedConsulted
Run, monitor and improveInformedAccountableConsultedConsultedConsultedResponsibleConsultedInformed

Business case and benefits

Director of water networks
A · Accountable
Leakage manager
R · Responsible
OT / telemetry engineer
C · Consulted
Solutions architect
C · Consulted
Technical PM
C · Consulted
Engineers
I · Informed
Security (NIS)
I · Informed
DPO
I · Informed

Target architecture and OT/IT boundary

Director of water networks
I · Informed
Leakage manager
C · Consulted
OT / telemetry engineer
C · Consulted
Solutions architect
A/R · Accountable and Responsible
Technical PM
C · Consulted
Engineers
C · Consulted
Security (NIS)
C · Consulted
DPO
I · Informed

NIS and OT security review

Director of water networks
I · Informed
Leakage manager
C · Consulted
OT / telemetry engineer
R · Responsible
Solutions architect
C · Consulted
Technical PM
I · Informed
Engineers
C · Consulted
Security (NIS)
A · Accountable
DPO
I · Informed

Telemetry, logger and meter data access

Director of water networks
I · Informed
Leakage manager
C · Consulted
OT / telemetry engineer
A/R · Accountable and Responsible
Solutions architect
C · Consulted
Technical PM
C · Consulted
Engineers
R · Responsible
Security (NIS)
C · Consulted
DPO
C · Consulted

DPIA for smart meter data

Director of water networks
I · Informed
Leakage manager
C · Consulted
OT / telemetry engineer
I · Informed
Solutions architect
C · Consulted
Technical PM
I · Informed
Engineers
I · Informed
Security (NIS)
C · Consulted
DPO
A/R · Accountable and Responsible

Build and integration with GIS and work management

Director of water networks
I · Informed
Leakage manager
C · Consulted
OT / telemetry engineer
C · Consulted
Solutions architect
A · Accountable
Technical PM
R · Responsible
Engineers
R · Responsible
Security (NIS)
C · Consulted
DPO
I · Informed

Model testing against repair history

Director of water networks
I · Informed
Leakage manager
A · Accountable
OT / telemetry engineer
C · Consulted
Solutions architect
C · Consulted
Technical PM
C · Consulted
Engineers
R · Responsible
Security (NIS)
I · Informed
DPO
I · Informed

Go-live decision

Director of water networks
A · Accountable
Leakage manager
R · Responsible
OT / telemetry engineer
C · Consulted
Solutions architect
C · Consulted
Technical PM
R · Responsible
Engineers
I · Informed
Security (NIS)
C · Consulted
DPO
C · Consulted

Run, monitor and improve

Director of water networks
I · Informed
Leakage manager
A · Accountable
OT / telemetry engineer
C · Consulted
Solutions architect
C · Consulted
Technical PM
C · Consulted
Engineers
R · Responsible
Security (NIS)
C · Consulted
DPO
I · Informed

R ResponsibleA AccountableC ConsultedI Informed· not involved

Where I'd fit

Option A

Lead the architecture

If there's no architect on the programme, I'd own the design end to end: the OT/IT boundary, the event pipeline that joins telemetry, loggers and meters, the integration with GIS and work management, and the gates. I'd work closely with your telemetry engineers and leakage team, who know the network better than anyone, and report to your programme board.

Option B

Support your architect

If your partner or your own team already has an architect, I'd take on the event pipeline, the agent and the work management integration under their lead, and give them a second senior opinion on OT security, data protection for meter data and whether the models are good enough to send crews out on.

Where this goes wrong

  1. Goes wrong

    Every signal raises its own job, and technicians visit the same leak three times.

    How the design avoids it

    Group events into one suspected leak by location and time window, with an ID on every event so retries never duplicate.

  2. Goes wrong

    Find rate is low, crews dig dry holes, and the field stops trusting the alerts.

    How the design avoids it

    Shadow mode first, a find-rate measure from day one, and technicians' findings fed back as labels.

  3. Goes wrong

    The pilot connects to telemetry outstations directly, and the OT team finds out afterwards.

    How the design avoids it

    Read from a historian replica through the DMZ, with telemetry engineers designing the connection.

  4. Goes wrong

    A repair is prioritised on volume alone, and a large interruption costs more than the leak.

    How the design avoids it

    Rank on volume, customer impact and risk together, and leave every valve and interruption decision with the control room.

  5. Goes wrong

    Smart meter data drifts into other uses nobody told customers about.

    How the design avoids it

    A DPIA, a stated purpose, DMA-level aggregation where possible, and access limited to the leakage team.

On the horizon

Changes already announced or under way that affect this design. More in the UK horizon scan.

  1. Outcome awaited
    Ofwat's leakage reporting methodology consultation (27 August to 9 October 2026) asked how to use more smart meter data, whether to keep night-flow calculation or move to a 24-hour water balance, and whether to phase changes in.

    Keep raw DMA and meter data, not just model outputs, so leakage can be recalculated under whichever method Ofwat decides.

  2. By 31 March 2027
    The interim target of a 20% cut in leakage from the 2017–18 baseline, then 30% by March 2032.

    Report the pilot DMAs' contribution in the same units as the company's leakage glidepath.

  3. Legislation to come
    The government plans a water bill to replace Ofwat with a single regulator covering economic, drinking water and environmental regulation, as set out in its January 2026 white paper.

    Keep evidence and audit trails in plain, regulator-neutral formats, so they serve whichever regulator asks.

  4. Late 2026–2028
    The Cyber Security and Resilience Bill tightens incident reporting for operators of essential services and brings medium and large managed service providers into regulation.

    Check whether your telemetry, logger or IT support providers are in scope, and update contracts and incident reporting to match.

  5. 2025–2030
    Water companies plan to fit about 10.4 million smart meters, reaching roughly half of homes and businesses by 2030.

    Design meter intake and storage for that growth, and keep the DPIA current as coverage rises.

Questions to ask your partner or vendor

Good answers are specific. Vague answers to any of these are worth following up.

Copy them into your next supplier meeting

  1. Can anything outside the OT network open a connection into telemetry or SCADA? How do you prove it can't?
  2. How will you measure find rate and missed bursts on our DMAs, not a benchmark?
  3. How do you stop one leak becoming three jobs when loggers, meters and night flow all spot it?
  4. Who decides a valve shut-off, and how are Guaranteed Standards and priority customers checked first?
  5. What is the lawful basis for using smart meter data, and who can see household-level readings?
  6. What does it cost to run per DMA per year, including logger and meter data?

Planning something like this?

Whether you need someone to lead the design or to support the architect you already have, let's talk it through.

Sam Naseem
Sam NaseemBirmingham, UK · UK & Europe
Available now