Building compliant and trustworthy AI

The UK’s wastewater network faces some of the most pressing infrastructure challenges of our time, with aging systems being put under increasing strain by population growth, river pollution and climate change. At the heart of tackling these issues is Arrow, an AI-driven solution developed by Atkins Realis to accelerate planning and optimise interventions. Grant Powell MBCS spoke to Josh Gay, the organisation’s Principal Data Scientist, to discuss the project’s recognition at the UK IT Industry Awards in the Data Science Project of the Year category, and how technology is reshaping the future of water management.

Please introduce yourself to our readers.

My name is Josh, and I’m a Principal Data Scientist at Atkins. I’ve been with the company for several years and currently manage a small team of data scientists. My experience spans multiple infrastructure sectors, including rail and aviation, but in recent years, I’ve focused on water. That’s where Arrow comes in. The project has been designed to address some of the UK’s most significant wastewater challenges.

What was the business challenge that Arrow was created to address?

The UK wastewater network faces multiple issues: flooding, river pollution and sewage spills into rivers, which is something we regularly see in the news. These problems are compounded by aging infrastructure, population growth, urban development and climate change, which brings more extreme rainfall events.

Water companies work in five year investment cycles known as asset management periods (AMPs). The current cycle, AMP8, involves billions of pounds of investment to tackle these challenges. But traditional planning methods — particularly hydraulic modelling — are slow and resource intensive. They require specialist skills that are in short supply, creating bottlenecks across the industry.

Arrow was designed to change that. By using AI and data-driven techniques, we can accelerate the planning of hydraulic interventions from months to weeks. This means water companies can deliver more interventions faster, while maintaining affordability for customers and meeting regulatory scrutiny.

What types of data are being leveraged?

Arrow starts with validated hydraulic models. These are digital representations of the wastewater network. These models include pipes, pumps, storm flows, storage tanks and sustainable drainage locations. They reflect real-world performance and form the foundation for designing interventions that reduce flooding or prevent spills into rivers.

How are advanced techniques being used?

Arrow uses evolutionary algorithms, specifically genetic algorithms, integrated with hydraulic models. We define objectives, such as reducing flooding in a town or minimising river spills, with the algorithm then running thousands of simulations — each testing different intervention designs such as upgrading pipes or adding storage tanks.
Over time, the system learns what works and what doesn’t, iteratively improving solutions against the objectives. This approach accelerates planning dramatically. Instead of spending months manually trialling options, engineers can set up the tool in a day or two, let it run simulations for a few days, and receive optimised designs that align with engineering principles.

How do you ensure data quality and integrity across large scale projects?

Data quality is critical. We use industry standard validated hydraulic models, which have already undergone rigorous checks to ensure they match real-world performance. Arrow builds on these trusted models, applying adjustments transparently and in an auditable way. Everything is regulator-ready, and we adhere to the same standards engineers have used for decades.

It’s worth noting that evolutionary algorithms differ from generative AI tools like ChatGPT. There’s no ‘black box’ inference here. Every output is grounded in engineering logic and traceable back to the hydraulic modelling software. This ensures full compliance, as well as engineer buy-in and trust.

What makes this project unique compared to other data-driven solutions?

Arrow’s uniqueness lies in its collaborative development. We didn’t just apply an algorithm to a problem; we worked closely with hydraulic modellers, engineers and clients to embed decades of best practice into the tool. Every output respects engineering principles, from pipe connection rules to depth requirements.

For you

Be part of something bigger, join BCS, The Chartered Institute for IT.

This engineer-first approach means the tool produces designs that feel familiar and trustworthy, while still challenging conventions when appropriate. In fact, there have been cases where engineers initially questioned a solution, only to later confirm it was a better approach than traditional methods.

That level of trust and adoption is only possible because we built Arrow with the end user in mind.

What measurable benefits do you expect for clients, and can you share any early success stories or pilot results?

The results so far have been impressive. In pilot projects, we’ve seen a 5-15% saving in capital expenditure on interventions, which is very significant when we consider that programs cost millions of pounds. We’ve also achieved a 50-80% reduction in operational expenditure, freeing up resources while delivering better outcomes.

Arrow has been successfully piloted with multiple regional water companies and is now being rolled out live across the UK. Feedback has been positive, and we’re continuing to scale.

How do you handle data privacy and regulatory compliance across different regions?

Data privacy is a top priority. Arrow offers both on-premises and cloud deployments, with strict tenancy controls. If a client requires all data to remain within their own environment, they can run the tool completely offline. For those needing more compute power, we provide an Azure-based option, but even that operates within secure, client-controlled tenancies. No data ever leaves the client’s environment.

What cybersecurity measures are in place for sensitive infrastructure data?

Security is built into every layer of Arrow. We use open-source components, avoid vendor lock-in, and ensure no external data sharing. Deployments are fully managed within client-controlled environments, whether on-premises or in the cloud. This approach guarantees transparency, control and compliance with regulatory standards.

Do you have any closing comments?

The Arrow project demonstrates the power of combining engineering expertise with advanced AI techniques. While generative AI dominates headlines, tools like evolutionary algorithms are quietly delivering tangible benefits in critical sectors. For the UK’s wastewater network, that means faster planning, lower costs and a healthier environment.
Finally, congratulations on achieving a Highly Commended status for your project. What are your thoughts on the awards?

The awards night was fantastic; well organised, informative and genuinely enjoyable. It was great to see so many talented people and innovative projects in the room. Being Highly Commended for Arrow was incredibly rewarding. It’s a recognition of the hard work my team has put in and the outcomes we’re starting to achieve. For me personally it highlights the importance of addressing water infrastructure challenges, which are among the toughest in the UK. Arrow is helping to support that effort, and this acknowledgment reinforces that we’re on the right track.