Your browser is out of date. Some of the content on this site will not work properly as a result.
Upgrade your browser for a faster, better, and safer web experience.

How water regulation needs to adapt to the emerging digital world

Published in The UK Water Report. Written by Martin Hurst – Sustainability First Associate

Introduction 
In the past, a reassuring thing about the water industry to a policy maker and regulator was the relatively slow pace of technological change. Some of the approaches to networks in the year 2000 would be recognisable to the Victorians. Many of the changes that there have been e.g. UV treatment for drinking water and smart meters, have been of a nature that they could be introduced incrementally.  The one possible exception – the automation of water and waste treatment – has been able to be funded through a combination of enhancement spend and company spending as an efficiency investment through the 5-year time profile of a price review. It has however become abundantly clear that this is no longer the case. Machine learning and its applications (pattern recognition, predictive maintenance, real time anomaly detection), digital twins, bioengineering and satellite-based approaches are all moving quickly, some at breathtaking pace.

It is essential both that regulation enables and encourages the emerging technologies, and that regulators go further and incorporate these tools in their own ways of working. Furthermore, demands on water are also moving rapidly through data centres, hydrogen production, climate change impacts and growth hubs etc.

Regulation needs to think about these combined challenges from multiple angles: technical (engineering and data systems); the efficiency and growth benefits; the environmental and consumer angles; and in how they can be absorbed into the practice of regulation.

My personal expertise is in policy and regulation, although I have been guided by people who understand the technology. But it is plain even to me that the current regulatory approaches are not well suited for this rapidly changing backdrop. Three years ago, Sustainability First judged that there was a lag of around 7 years between valuable new technologies arriving and regulation accepting them: more in cases where the response of regulators/legislators was to create a series of pilots rather than mainstreaming new practice.

While this delay may have been annoying in the past, addressing it is now mission critical given the pace of change.   A technological development with a clear evidence base produced in say late 2027 cannot wait for the methodology statement and Water Industry National Environment Plan (WINEP) for the 2034 price review. Indeed, in some cases it can’t even wait for 2029. The importance of taking on board digital and other changes was flagged in the Corry review and parts of the Cunliffe review.

Ofwat are aware of the issues, and have tried to start addressing them, for example through their innovation fund and a recent publication pledging “regulatory sandboxes” and other proposals. But this remains far too short term and incremental and it doesn’t appear to extend to environmental regulation.  In a previous Water Report Dr Melvyn Weeks also correctly argued that regulators must not only reform how they support and incentivise machine learning in the industry but also adopt it themselves as part of their regulatory toolkit.

What needs to change – policy 
So, what should be done? The sector needs a proper conversation here, but a few suggestions:

There is a case for what is called a Futures approach similar to an approach taken in energy sector and on water resources. This explores questions such as where technology might be in 5, 10- and 25-years’ time,  what difference might it make to the sector and its stakeholders, and creates a number of “destination statements”. We could then work backwards to judge when these technologies might be ready for industry use—allowing time for proof‑of‑concept to manage risks—and decide what we should be doing now to prepare for or develop them. Some of this thinking has already been done, though more in the abstract than for the particular needs and applications in the water industry.

This doesn’t need to be a lengthy process and could be augmented by building regulator capability and understanding for example through the Cunliffe review recommendation that the new regulator has a chief engineer which should be expanded to cover a chief technology lead.  (I accept of course that the industry will need to lead on much of this: although economic literature states that monopoly industries will underinvest in R and D there is some good stuff being done.)

This approach needs to be supported with an ‘observatory’ to map changes as they arise and identify the data needs in advance to ensure telemetry etc is optimised for the uses it can be put to. It also requires something close to a systemic change in regulation and regulators.   A regulatory system which clamps down on baseline totex spend will tend to under-incentivise operational spend, such as some forms of improved telemetry, relative to major capital spend – despite the absolute criticality of telemetry investment to maximising the impact of machine learning. Capital maintenance regimes need to stop being backward looking and incorporate the business case for timing major maintenance with development of relevant technology and the opportunities the new technologies give for smarter approaches.

An uncritical application of fixed 5-year price review cycles – with inflexible spending agreements and incentive regimes/price control deliverables fixed over 5 years – is almost the worst approach to rapidly changing technologies. And this is compounded by a rigid and risk averse process based regulatory approach to the environment, most notably on wastewater, with the kind of very high evidential bar to anything new which is an anathema to innovation and technological development. We need to move to more flexible regimes, with for example greater use of reopeners both in Ofwat and the WINEP, so companies can make reasonable cases to change the quantum and form of their spending: in some cases, to reduce steel and concrete spend for less spending on more efficient alternatives. We also need to align regulatory incentive mechanisms with the pace of technological change.

There is also a question of how customer regulation works in a machine learning age. Whose job is it for example to ensure that chatbots don’t reduce company costs but increase hassle for consumers? Who needs to ensure that digitally disenfranchised customers don’t lose out? How should Ofwat’s customer service incentives adapt so we reap the genuine benefits of AI for customers? And what is needed to augment a system of oversight which is based around slow moving surveys, fixed KPIs and relying on people to complain.

Industry suggestions 
Industry and other colleagues have given me some examples of what might be possible:

Machine learning will be a major driver of the next phase of automation, both of key treatment sites and of networks. AI works particularly well where there is good quality, consistent machine-readable data. Constrained (‘deterministic’) systems such as engineered assets are particularly well suited to this. This can also give early warning of potential failures, before they occur: if improved telemetry means that there is data at different points across the system recording rates of flow, then machine learning  models will spot when something odd is happening, and allow intervention before there is a major burst.

We can now get much better weather forecasting of localised rainfall such as the new ‘2km grid square’ approach enabled by the Met Office supercomputer. With data on river flow etc and a ‘digital twin’ model of a sewage works, we can prepare much better for the kind of sudden rainfall which can result in sewage overflows.

Beyond network monitoring, there are already other applications where the technology is sufficiently mature that the question is not what to do, but how to regulate it. AI- assisted condition assessment with computer vision and pipe inspection cameras can optimise capital maintenance. Where there is Smart meter data, leakage, demand profiles and even vulnerability can be mapped and issues flagged. (If data sharing is the concern, then readers should look at our website for the smart meter data sharing protocols we worked up with stakeholders for energy.)

Looking further ahead, it seems likely that developments in real time monitoring (the ‘section 82 continuous water quality monitors being an early example) and data interrogation could extend from engineering networks to natural catchments – answering the more difficult challenges of multiple interacting variables and sparser, noisier (‘stochastic’), data. This would allow the creation of accurate base models for those catchments which can then be calibrated for given types of climate change or short-term weather impacts. In that case the health of water bodies could be monitored in real time (including using satellite data) and if that health started to deteriorate suddenly then intervention could be quickly taken. If the problem was, say, an agricultural spill then it could be stopped before the worst impacts materialise, rather than waiting until we see dead fish. There is obvious potential here for regulators to move from expensive and carbon intensive requirements for storm tanks, to outcome-based regulation targeting the health of the water body.  This could also be the evidence base which allows trading of environmental outcomes, advocated strongly by the Corry review, to take off.

Equally, rather than relying on mechanistic ‘star ratings’ based on a series of process measures, companies’ performance could be judged by machine learnt assessment of their environmental performance, and a regulator could move towards risk-based regulation, taking the regulatory cosh off companies who have established a successful track record.

Finally, a point which Dr Weeks made in his earlier article, Ofwat rely heavily on data intensive econometric models. These are essentially backward looking (using historic datasets). New AI type approaches can augment this with ‘learning’ and forward-looking approaches using machine learnings ability to spot emerging patters and to update estimates. Care is needed: it still seems the case that the existing AI tools here appear to have some systematic biases. But I find it hard to believe that PR29 needs to be run on the same methodologies as previous price reviews. Equally AI could negate the need for much more intrusive regulation under the supervisory approach.

Risk 
In all of this a proper discussion about risk appetite is essential. We should be highly risk averse to anything which might, if it goes wrong, lead to supply interruptions. And there is evidence that machine learning engines currently have a systematic optimism bias. But not rising to the challenge is itself a huge risk, leading to excessive bill increases and underperformance against the key outcomes of water resource resilience and an improving and more resilient water environment.  Companies and regulators also need to consider potential hostile uses of new technologies. In the shorter term, AI assisted hostile cyber-attacks could identify and exploit vulnerabilities in operational systems. Water infrastructure may well be a key target: either from hostile states, or for ransomware. And in the longer term, quantum computing and AI could lead to systematic threats to encryption technology and therefore to customer data: something already flagged by the national cyber security centre. As more key systems are automated  “no- digital” backs ups become essential.

Conclusion 
To sum up: technology is advancing so fast that traditional regulation risks missing major opportunities and locking in capital‑heavy, process‑driven approaches instead of operational and outcome‑based ones. That would mean losing the chance to improve the environment and cut bills. Companies can’t wait for regulation; they must innovate, and regulators should challenge any complacency that comes from monopoly positions.

We need far more flexible regulation, including greater freedom to shift spending within price reviews. Wider use of reopeners seems the obvious route. We should also look further ahead, using futures thinking rather than relying on pilots and incremental tweaks. New technology could be the key to moving away from opaque econometrics and wasteful process regulation towards learning‑based, outcome‑focused approaches.

We need an open conversation about all this. At Sustainability First, we plan to start that discussion—let me know if you’d like to be involved.

Email: communications@sustainabilityfirst.org.uk

Fair for the Future ofwat Regulation Water Expert viewpoints

Want to find out how we can help your organisation?

Get in touch

Search our publications