AI for Water Utilities: Moving Beyond Prediction to Consequence-Driven Asset Management

The next era of water utility management demands more than predictive AI. Real transformation depends on consequence-based decision frameworks that link AI insight to regulatory, customer and business outcomes. That is where value from the AMP8 investment cycle is won or lost.

How Can Water Utilities Turn AI Predictions into Defensible Decisions?

Water utilities turn predictions into defensible decisions by adding a governed decision layer between the model and the intervention. That layer maps each predicted failure to its customer impact, its regulatory consequence and its cost, then ranks the options on those terms rather than on risk score alone. Hexaware builds this layer for utilities using artificial intelligence and agentic AI services grounded in a trusted data foundation.

Introduction

Water companies in England and Wales have entered the largest investment programme since privatisation. The 2024 price review (PR24) determinations set by Ofwat, the Water Services Regulation Authority, provide the sector with what Ofwat calls a record £104 billion of investment for the 2025 to 2030 period, as confirmed in its Water Company Performance Report 2024-25, published in October 2025. A large share of that money, and much of the sector’s AI spend alongside it, is going into prediction. Models now forecast which main will burst, which pump will fail and which overflow will spill. Far less attention is going to the capability that converts a prediction into a decision somebody can defend.

That gap is about to become expensive. In its January 2026 white paper, A new vision for water, the Department for Environment, Food and Rural Affairs (Defra) confirmed it will abolish Ofwat and create a single integrated water regulator, bringing together the relevant water system functions of Ofwat, the Drinking Water Inspectorate (DWI), the Environment Agency and Natural England into one body. Defra states that the new regulator will take a supervisory approach targeted to the specific needs of each company, with a sharper focus on long-term financial resilience, asset health and environmental performance. Ofwat, in parallel, is consulting on a new licence condition covering asset management competence.

Water utility management is being reframed in front of the sector. The regulator will no longer ask only what happened to your assets. It will ask how you decided what to do about them, and asset management maturity is now the evidence base for that answer.

The Problem: Prediction Without Consequence

The Limits of Predictive AI

Over the past decade, water utilities have invested heavily in models that forecast asset failure, predict pollution incidents and optimise maintenance schedules. Those tools are no longer differentiators. They are the baseline for modern water infrastructure asset management.

The performance record tells a sobering story. Ofwat reports that the sector committed to a 30% reduction in pollution incidents across the 2020 to 2025 period, yet actual performance deteriorated by 27%. Ofwat notes that companies achieved a 15% reduction in the first three years. The increase in the final two years erased it.

The environmental data is worse. According to the Environment Agency’s Water and sewerage companies in England: environmental performance report for 2024, published on 23 October 2025, serious pollution incidents rose 60% in a single year, from 47 in 2023 to 75 in 2024. The Environment Agency records total pollution incidents from sewerage and water supply assets at 2,801, a 29% increase on the 2,174 recorded in 2023, and the highest figure since 2011. The same report attributes 61 of those 75 serious incidents, or 81%, to the assets of three companies: Thames Water with 33, Southern Water with 15 and Yorkshire Water with 13.

Ofwat states that underperformance payments returned more than £260 million to customers in 2024-25 alone, and more than £700 million across the 2020 to 2025 period. Ofwat also reports that, as of October 2025, five concluded enforcement investigations had secured more than £240 million in financial redress. Those are the direct financial consequences of decisions that failed to deliver for customers and the environment.

Why Prediction Alone Falls Short

The core issue is not a shortage of predictions. Most utilities can already anticipate which main is likely to burst or which pump is degrading. What is missing is the ability to convert that anticipation into a consequence-aware decision: which intervention, when, at what cost, and with what impact on customers, compliance and business value.

Consider where the incidents actually originated. The Environment Agency’s 2024 breakdown attributes incidents to foul sewers (865), pumping stations (693), sewage treatment works (647) and rising mains (176), among other asset classes. It records 2,469 incidents from sewerage assets and 332 from water supply assets, the latter up 22% on the 272 logged in 2023.

Each of those asset classes carries a very different consequence profile. A rising main failure and a distribution main failure are not comparable events, however similar their condition scores look. A single ranked risk register cannot express that difference. Water utility decision making needs a layer that can.

The Regulatory Shift: AMP8 and Asset Management Maturity

AMP8: Raising the Bar for Water Utilities

AMP8, the eighth Asset Management Plan period, runs from 2025 to 2030 and represents the most ambitious investment cycle in the sector’s history. Defra describes the £104 billion programme that the sector will deliver between 2025 and 2030 as a down payment on transformative change, and as a break from what it calls the underinvestment of the past.

Investment on that scale attracts scrutiny on the same scale. In the same white paper, Defra commits to funding being directed appropriately to maintain assets, with separate allowances for capital maintenance, operating expenditure and enhancement capital expenditure at future price reviews. Defra also commits to working with the new regulator to develop forward-looking asset health metrics, building on Ofwat’s plans for a data-gathering exercise to provide a snapshot of asset conditions, and to publishing a Transition Plan during 2026. AMP8 asset management is therefore not only about spending the money. It is about evidencing the logic behind it.

Asset Management Maturity as a Licence Condition

Ofwat has moved from consultation to draft licence text, and the detail has shifted in ways that matter to anyone building an AI roadmap.

In its Proposal to improve asset management maturity of water companies: conclusion and decision document, published in November 2025, Ofwat confirmed it would not require every company to attain and hold certification to ISO 55001:2024, the asset management standard published by the International Organization for Standardization (ISO). Instead, Ofwat will require companies to demonstrate competency in a way that is appropriate and proportionate to their organisation, treating ISO 55001:2024 as best practice. Companies that do not certify must have their competence assessed by an appropriate independent practitioner. Ofwat also confirmed in that document that it extended the transition period from one year to two, that ongoing monitoring will be reported through the Annual Performance Report (APR) process, and that the sector-wide Asset Management Maturity Assessment (AMMA) sits outside the licence requirement and will be delivered in 2026 and in the second year of each subsequent AMP.

Ofwat’s May 2026 consultation under sections 13 and 12A of the Water Industry Act 1991 puts that decision into draft licence wording. It proposes removing existing Condition L and introducing a new condition on asset management competence. Ofwat states that by the relevant date in each financial year, companies will need to submit evidence demonstrating a competent level of asset management maturity. Holding a valid ISO 55001:2024 certification and submitting the most recent surveillance audit results is one compliant route. Ofwat is explicit that the condition does not prescribe a single method, and that companies may adopt different approaches provided they can demonstrate credible and independent assurance that asset management capability is effective in practice.

That flexibility is generous on paper and demanding in practice. A company that cannot trace how a decision was reached will struggle to produce credible assurance evidence, whichever route it chooses.

The Missing Layer: Consequence-Based Decision Making

What Is the Consequence Layer?

We call the missing capability the Consequence Layer. It is a governed, water-specific model of how the business actually hangs together: assets, network zones, treatment processes, work orders and crews, customers and Priority Services Register (PSR) status, cost, and the performance commitments and outcome delivery incentives (ODIs) that attach to each of them.

Given a proposed intervention, or a proposed deferral, the layer can trace what moves. Rather than stopping at what will happen, it answers what should be done and why. Hexaware has explored this first-hand with a UK water utility, connecting information across assets, operations, networks, field activity, customers and regulatory outcomes.

Example: From Asset Failure to Customer Impact

Consider a rising main, the pressurised pipe that carries wastewater uphill from a pumping station. From an asset perspective it has a condition grade, a criticality score, a location and a maintenance history. For customers, its failure means sewage in gardens or homes and a call to the contact centre. For the regulator, it is a pollution incident, an internal sewer flooding count and an ODI payment.

The asset has not changed. The business impact has. Each of those views typically lives in a different system, owned by a different team, and reconciled manually if at all. The Consequence Layer evaluates them together:

• The likelihood and severity of customer disruption
• The regulatory penalty attached to a service failure
• The cost and timing of the intervention
• The effect on long-term asset performance management

Why Consequence-Based Management Matters

Regulatory direction is converging on exactly this capability. Defra ties the new regulator’s supervisory model to long-term financial resilience, asset health and environmental performance. Ofwat’s proposed licence condition requires demonstrable competence rather than a stated intention. A water utility management strategy built only on predictive accuracy satisfies neither.

There is an operational argument too. Ofwat’s performance report shows the sector improving where outcomes were clearly defined and owned. Ofwat records PSR reach growing from 2% to 12.8% across the period, drinking water compliance holding at 99.97%, leakage down 43% since privatisation, and internal sewer flooding improving, with two companies achieving reductions of 70% or more. Progress follows clarity about consequence.

AI Agents and the Risk of Context-Free Automation

The Rise of Agentic AI in Utilities

Utilities are now deploying autonomous agents across business functions. The hard part is not getting each agent to work on its own. It is getting them to understand the same business situation.

According to Deloitte Insights, published in October 2025, nearly 40% of utility control rooms are expected to use AI by 2027. Deloitte also notes that guidance from the North American Electric Reliability Corporation (NERC) emphasises that AI should serve as a decision-support tool rather than an autonomous controller, and that human oversight remains central to strong governance as adoption broadens. The direction of travel is clear. So is the guardrail.

Project Success Hinges on Context

The governance risk is now quantified. According to Cybersecurity Insiders’ 2026 CISO AI Risk Report, a survey of 235 large-enterprise CISOs and CIOs, 92% of organisations lack full visibility into their AI agent identities, and 95% doubt they could detect or contain a compromised agent. The same report finds that 86% do not enforce access policies for AI identities, and only 16% govern that access effectively. The gaps in agent governance, in other words, are the kind that only reveal themselves once something has already gone wrong.

For a regulated water utility, that gap is not academic. An agent that can raise a work order, defer a job or reprioritise a crew is making a regulatory decision, whether or not anybody labelled it as one. Without visibility into what each agent can access, and a shared model of consequence, agents optimise locally and create exposure globally.

Building Consequence-Driven Asset Management: A Practical Approach

Integrating AI With Business Context

To unlock the full value of AI in water utilities, four moves matter more than model selection.

• Map asset data to business outcomes: Link asset condition and performance data to customer experience, regulatory commitments and financial objectives. This is a data strategy and roadmap exercise before it is an AI exercise.
• Embed regulatory logic into decision models: Performance commitments, ODIs and licence obligations belong inside the model, not in a review conducted after the fact.
• Prioritise interventions by consequence: Move past the risk score. Rank options by real-world impact on customers, compliance and cost.
• Enable transparent, auditable decisions: Record the rationale, the alternatives considered and the trade-offs accepted. That record is what an independent assessor will ask to see.

Hexaware supports each step with data and analytics services, cloud transformation services for the underlying platform, enterprise automation services built on Tensai®, and AI-powered platforms including Amaze® for cloud transformation and Agentverse™ for governed agent orchestration. The full portfolio sits across Hexaware’s IT services and solutions practice.

Use Case: Low-Pressure Complaints

On a Monday morning, the contact centre logs a cluster of low-pressure complaints from one district metered area (DMA). In a utility with a Consequence Layer, that cluster is a query, not a ticket.

The asset view shows the pressure reducing valve feeding the zone was flagged by the condition model three weeks earlier as deteriorating. The customer view sizes the exposure: several thousand properties, a proportion of them on the PSR. The regulatory view attaches the consequences, including customer contacts and C-MeX, Ofwat’s Customer Measure of Experience, today, plus a potential supply interruption performance commitment breach if the valve fails outright.

Now the decision maker has options with consequences attached rather than a symptom to triage. Expedite the valve repair, with the cost and the knock-on effect on planned work stated openly. Or hold, with the customer, service and regulatory exposure of holding expressed in the same units as the cost of acting. Both are decisions. Both are defensible, because the reasoning is traceable.

Conclusion

AMP8 is a turning point, and not only because of the money. The regulatory question is changing from what happened to your assets to how you decided what to do about them. Prediction cannot answer that. A consequence layer can.

The highest-risk asset on your model is often not the asset you should fix first. If your AI cannot tell you why, it is not yet ready for AMP8. The future of AI for water utilities is not about teaching machines to understand data. It is about teaching machines to understand the utility.