For decades, planned preventive maintenance (PPM) in the water sector has operated on a simple principle: maintain assets on a fixed schedule, regardless of their actual condition, risk profile, or operational context. It is a model built for predictability, not performance - and in an era defined by ageing infrastructure, tightening regulatory scrutiny, and the exponential growth of network data, it is showing its limitations.

The question water companies are now asking is not whether to move beyond fixed-schedule maintenance. The question is how quickly they can build the capability to do it intelligently.

Dynamic PPM or risk-based maintenance planning - represents the next evolution of asset management in the water sector. Powered by AI and machine learning, it replaces static calendar-driven cycles with continuously updated, risk-prioritised maintenance decisions. It draws on live telemetry, historical failure data, operational context, and environmental inputs to ensure the right work gets done on the right asset at the right time.

This article examines the strategic case for AI-driven risk-based maintenance planning, the technical foundations required to deliver it at scale, and the practical steps water companies can take to begin the transition.

The Limits of Fixed-Schedule PPM in Modern Water Networks

Traditional PPM frameworks were designed for a simpler operational environment. Assets were inspected or serviced at fixed intervals - annually, quarterly, or on a mileage or usage basis - and maintenance teams planned their workloads accordingly. The approach has genuine strengths: it is predictable, auditable, and easy to resource.

But it carries a fundamental flaw. It treats all assets as equivalent, regardless of their actual risk exposure, deterioration trajectory, or operational criticality. A pump that serves a major distribution trunk main is maintained on the same cycle as one servicing a low-priority secondary network. A sewer with documented blockage history is inspected no more frequently than one with a clean record. The result is a systematic misallocation of maintenance resource - too much work on assets that do not need it, and not enough on those that do.

In practice, this creates three well-documented operational problems:

1. Reactive maintenance spikes, as assets that should have been prioritised fail unexpectedly, triggering costly emergency callouts and service disruptions.

2. Regulatory exposure, particularly around pollution incidents and leakage performance, where missed or poorly timed maintenance contributes directly to category-one and category-two events.

3. Capital inefficiency, as maintenance budgets are spread evenly across the asset base rather than concentrated where they generate the greatest risk reduction.

Industry context:

Water companies in England and Wales collectively manage over 300,000 kilometres of water mains and 340,000 kilometres of sewers. Even marginal improvements in maintenance prioritisation - redirecting a fraction of annual maintenance spend to the highest-risk assets - translate into material reductions in pollution incidents, service failures, and regulatory penalties.

What Risk-Based Maintenance Planning Actually Means

Risk-based maintenance (RBM) is not a new concept. The water sector, like oil and gas, power generation, and rail, has discussed the principles for years. What has changed is the practical feasibility of implementing it at scale - and that change is being driven by AI.

At its core, a risk-based maintenance model answers a different question than traditional PPM. Rather than asking when an asset is next due for maintenance, it asks: which assets in my network present the greatest risk right now, and what action would most efficiently reduce that risk?

Answering that question requires four capabilities working together:

1. Continuous risk scoring

Every asset in the network - pump, valve, pipe, treatment unit, telemetry node - is assigned a dynamic risk score that updates in real time as new data flows in. The score reflects the probability of failure, the consequence of failure, the asset's current condition, and its operational context. Assets are ranked and re-ranked continuously, not once a year.

2. Multi-source data integration

Risk scores are only as good as the data that underpins them. Effective RBM draws on asset register data, maintenance history, telemetry readings, weather and environmental inputs, operational event logs, and - increasingly - smart meter data from the distribution network. Fragmented or siloed data is one of the most common barriers to implementation.

3. Workload optimisation

Risk prioritisation must be balanced against practical constraints: crew availability, geographic clustering of work, permit and access requirements, supply chain lead times, and the interaction between planned and reactive work. AI-driven workload optimisation models can solve this scheduling problem at a scale and complexity that manual planning simply cannot match.

4. Continuous learning

As maintenance actions are completed and their outcomes recorded, the model learns. It identifies which interventions successfully reduce failure probability, where its predictions were wrong and why, and how its risk scoring should be updated. Over time, the model becomes progressively more accurate.

The Data Foundations: Why This Is Harder Than It Looks

For most water companies, the journey to risk-based maintenance does not begin with AI. It begins with data.

The sector is sitting on vast volumes of operational and asset data: SCADA and telemetry systems, maintenance management platforms, geographic information systems, customer and billing data, environmental monitoring feeds, and an expanding base of smart meter deployments. The challenge is not a lack of data - it is that this data typically exists in disconnected silos, maintained by different teams, stored in incompatible formats, and updated on different cycles.

Without a unified operational data foundation, risk-based maintenance planning cannot function. An AI model that scores asset risk cannot do so coherently if the asset register does not align with the maintenance system, or if telemetry data from one network zone is unavailable to the model scoring assets in another.

Building that foundation typically involves three workstreams:

1. Data consolidation and harmonisation: Bringing together asset, maintenance, operational, and environmental data into a single accessible layer, resolving inconsistencies, duplicates, and gaps.

2. Real-time data pipeline development: Establishing the infrastructure for live telemetry and operational event data to feed AI models continuously rather than in periodic batch updates.

3. Data governance and quality assurance: Implementing the processes and tooling to ensure data quality is maintained over time, with clear ownership and escalation paths for data issues.

This investment in data infrastructure is not a precursor to AI - it is part of the AI programme. The two workstreams should be designed together from the outset.

VE3 perspective:

A recurring finding across water sector engagements is that organisations underestimate the data harmonisation challenge and overestimate how quickly AI models can be trained on imperfect data. VE3 recommends a parallel-track approach: begin building the data consolidation layer immediately, while running AI model development on the cleanest available data subset - then progressively expand coverage as data quality improves.

AI Techniques That Power Risk-Based Maintenance

The AI capability stack for dynamic PPM draws on several techniques, applied at different layers of the problem:

Predictive failure modelling

Machine learning models trained on historical maintenance records and telemetry data identify the early indicators of asset deterioration and likely failure. Gradient-boosted tree models and deep learning architectures have both demonstrated strong performance in water network applications. The choice of technique depends on data availability, the asset class in question, and the required explainability of outputs.

Survival analysis and time-to-failure estimation

Rather than classifying assets as likely to fail or not, survival analysis models estimate the probability distribution of remaining useful life. This provides maintenance planners with a richer, probabilistic view of risk - enabling confident decisions about assets with moderate risk profiles, not just those already approaching failure.

Anomaly detection

Real-time anomaly detection models monitor telemetry streams for deviations from expected operational patterns, flagging potential developing faults before they appear in explicit failure data. This is particularly valuable in networks where failure events are relatively infrequent, making supervised training data sparse.

Optimisation algorithms

Once risk scores and maintenance recommendations are generated, optimisation algorithms - including constraint satisfaction solvers and reinforcement learning models - balance risk reduction against operational and resource constraints to produce practical, deliverable work schedules.

Explainable AI (XAI) for regulatory confidence

Water companies operate in a regulated environment with accountability obligations. Maintenance decisions need to be explainable - to internal leadership, to Ofwat, and increasingly to boards who must understand the basis on which asset risk is being managed. XAI techniques, including SHAP (SHapley Additive exPlanations) values and decision-tree approximations of complex models, are increasingly standard components of production AI systems in the sector.

Integration with ISO 55000 Asset Management Frameworks

For water companies working towards ISO 55000 compliance, AI-driven risk-based maintenance planning is not merely operationally advantageous - it is architecturally aligned with the standard's requirements.

ISO 55000 demands that organisations demonstrate a systematic, risk-informed approach to asset management decision-making. It requires documented evidence that maintenance priorities are set based on a clear understanding of asset risk, consequence of failure, and organisational objectives. Fixed-schedule PPM, applied uniformly regardless of risk, does not satisfy this requirement in any meaningful sense.

By contrast, a well-implemented dynamic PPM system provides:

1. Continuous, auditable risk scoring for every asset in the network, updated in real time and traceable to underlying data.

2. Documented decision logic for maintenance prioritisation, explainable to auditors and regulators.

3. A closed feedback loop between maintenance actions and risk outcomes, enabling continuous improvement and demonstrating the organisation is actively learning from its asset management decisions.

4. Integration between planned preventive work, reactive maintenance, and capital renewal programmes - giving a whole-life view of asset risk management.

For companies facing two to three year timelines for ISO 55000 compliance, beginning the shift to AI-driven risk-based maintenance now is both a practical and a strategic imperative. The data infrastructure and governance frameworks required for dynamic PPM are the same foundations that underpin a credible ISO 55000 programme.

Blockage Prediction and Prevention: A Worked Example

One of the clearest demonstrations of AI-driven risk-based maintenance in practice is in wastewater network blockage prediction and prevention - an area with direct links to both operational efficiency and regulatory performance.

Traditional sewer maintenance follows fixed jetting and CCTV inspection cycles. High-risk sewers in dense urban areas are typically maintained every one to three years; lower-risk assets on longer cycles. This approach generates a predictable maintenance programme but does not reflect the actual dynamics of blockage risk, which varies with factors including catchment land use, fat-oil-and-grease (FOG) discharge patterns, connectivity to food service premises, pipe gradient and condition, and flow volumes.

AI-driven blockage risk models draw on these factors - along with historical blockage event data, operational logs, weather data, and where available, real-time flow monitoring - to assign dynamic risk scores to individual sewer sections. The result is a continuously updated map of network risk that maintenance teams can use to target jetting and inspection resources on the sections most likely to block, rather than those simply next in the calendar cycle.

Evidence from deployments across major UK water utilities has demonstrated:

1. Reductions in total blockage incidents of 20 to 35 per cent in networks where predictive models are used to inform maintenance scheduling.

2. Significant reductions in the proportion of blockages that escalate to pollution events, as higher-risk sections receive earlier intervention.

3. Maintenance efficiency gains as jetting resources are concentrated on genuinely high-risk sections, reducing unnecessary work on assets with consistently low blockage probability.

The same AI capability - risk-scored, continuously updated, operationally integrated - applies across the broader wastewater asset base: pumping stations, storm overflows, screening equipment, and treatment processes.

Leakage Detection and Clean Water Network Optimisation

On the clean water side of the network, the parallel capability is AI-driven leakage detection and distribution network optimisation.

Smart meter rollouts are transforming the data landscape for UK water companies. Thousands of consumption readings per day, per customer, combined with pressure and flow monitoring across the distribution network, create the conditions for highly granular leakage detection - if the data can be processed effectively.

AI models trained on smart meter and pressure data can identify the characteristic signatures of developing mains bursts, service pipe leaks, and background leakage at a spatial resolution and speed that far exceeds traditional district metered area (DMA) monitoring. Early identification means faster repair, less water lost, and reduced risk of the ground movement and property damage that accompanies significant burst events.

Beyond leakage detection, the same data infrastructure supports dynamic pressure management - continuously optimising network pressure to minimise both burst probability and background leakage, without compromising customer service levels.

For companies with active smart meter programmes, the investment in AI-ready data infrastructure should be planned in parallel with the meter deployment. The data will be available; the question is whether the analytical capability exists to use it.

Whole-Network Visibility: From Catchment to Customer Tap

The full strategic value of AI-driven risk-based maintenance is realised not when it is applied to individual asset classes in isolation, but when it provides a unified, catchment-wide view of network risk - integrating wastewater and clean water assets, operational telemetry, environmental inputs, and planned work programmes into a single operational picture.

This whole-network visibility enables a qualitatively different kind of decision-making. Rather than managing the wastewater network separately from the clean water network, and both separately from capital programmes and regulatory reporting, operations leadership gains a real-time view of where risk is concentrating across the entire system - and how different interventions would reduce it.

Weather data integration is particularly valuable in this context. Rainfall events drive combined sewer overflow (CSO) activity, elevate blockage risk, and affect treatment plant performance. AI models that incorporate weather forecasts alongside network telemetry can anticipate pressure points and pre-position maintenance resources before conditions deteriorate - moving from reactive to genuinely predictive operations.

The organisational benefit is equally significant. Whole-network visibility reduces the coordination overhead between operational teams, enables earlier identification of systemic issues that cross departmental boundaries, and provides the evidence base for more credible regulatory submissions.

Implementation Considerations: What Water Companies Need to Know

The shift to AI-driven risk-based maintenance planning is a significant operational and organisational change, not simply a technology deployment. Companies that approach it as the latter consistently underperform against those that treat it as a strategic programme with technology as an enabler.

Key implementation considerations include:

Stakeholder alignment and change management

Maintenance planners, field engineers, and operations managers need to understand how AI-generated risk scores are produced, and trust them sufficiently to act on them. This requires investment in communication, training, and - critically - a period of parallel running during which AI recommendations can be validated against experienced human judgement before full handover.

Phased deployment and measurable value

Risk-based maintenance programmes deliver the greatest adoption success when they are introduced in phases, with clear KPIs at each stage and tangible operational wins that build internal confidence. Beginning with a well-bounded, data-rich asset class - blockage prediction in a specific catchment, for example - allows the programme to demonstrate value quickly before scaling.

Integration with existing systems

AI-generated maintenance recommendations need to flow into existing work management, scheduling, and field mobility systems to be operationally effective. Integration with platforms such as IBM Maximo, SAP PM, or sector-specific WAMS is typically a significant technical workstream in its own right.

Regulatory and cyber security requirements

Water companies operate under significant regulatory oversight, and AI systems that influence maintenance decisions on critical national infrastructure are subject to scrutiny. Data security, model governance, auditability, and supplier vetting requirements need to be addressed from the outset of programme design.

VE3 experience in regulated environments:

VE3 has delivered AI and data programmes across regulated infrastructure sectors, with direct experience of the security, governance, and auditability requirements that apply to operational technology environments. Our approach includes rigorous data security design, model explainability as a standard deliverable, and a supplier assurance framework aligned to public sector and regulated industry standards.

How VE3 Approaches Risk-Based Maintenance for Water Companies

VE3 Global brings together enterprise AI capability, operational technology experience, and sector-specific knowledge to help water companies build the data and AI foundations for dynamic PPM.

Our approach to risk-based maintenance planning programmes is structured around three integrated workstreams:

1. Consolidating and harmonising asset, maintenance, telemetry, and environmental data into a unified, AI-ready data layer - resolving fragmentation and establishing the governance frameworks to maintain data quality over time.: Data foundation

2. Building and validating predictive failure, anomaly detection, and workload optimisation models on real network data, with explainable outputs designed to meet regulatory and internal governance requirements.: AI model development and deployment

3. Integrating AI recommendations into existing work management systems, supporting the change management process with maintenance and operations teams, and establishing the monitoring and continuous improvement frameworks that keep the programme delivering value over time.

‍Operational integration and change enablement

VE3's experience with blockage prediction and prevention programmes at major UK water utilities provides directly applicable proof points - including validated model performance data, integration architectures, and lessons learned from operational deployment - that can accelerate programme development and reduce delivery risk for new clients.

Conclusion: The Strategic Case for Acting Now

Dynamic PPM and AI-driven risk-based maintenance planning represent a structural shift in how water companies manage their asset base. The technology is mature. The sector data is increasingly available. The regulatory and commercial pressure to deliver better outcomes with constrained maintenance budgets is intensifying.

The organisations that will lead on pollution prevention, leakage reduction, ISO 55000 compliance, and operational efficiency over the next AMP cycle are those investing now in the data foundations and AI capabilities that make risk-based maintenance possible at scale.

The fixed-schedule model served its purpose. The infrastructure exists to replace it with something demonstrably more effective. The question for water company leadership is no longer whether to make this transition - it is how to do it in a way that delivers measurable value quickly and builds towards a genuinely intelligent, whole-network operational capability.

About VE3

VE3 is a UK-based technology and enterprise AI consultancy, partnering with water companies, regulated utilities, and public sector organisations to deliver AI, data, and digital transformation programmes that create measurable operational and commercial value. With offices in London and Pune, VE3 combines deep sector knowledge with cutting-edge AI capability to help clients navigate the full journey from data strategy to production AI deployment. Get in touch now.