Entering a new realm of project assurance

Harnessing AI to reduce ambiguity and deliver better value for clients

Verifying requirements at the start of a large-scale project is a manual process that can be time-consuming and repetitive. Ranjit Davis, global discipline lead for systems engineering and assurance, explains how artificial intelligence (AI) is being used responsibly to help engineers bring greater value to project assurance and delivery.

Setting out the requirements of a major project is an enormous endeavour that often involves thousands of pages of requirements and technical specifications. To keep projects on track there must be absolute clarity about what needs to be delivered. Any ambiguity can be extremely costly, leading to rework and delays.

To avoid any setbacks, engineering companies traditionally go through a painstaking process of manually verifying project requirements. But this is very time-consuming and inefficient for today’s complex mega projects, which can include thousands of requirements spread across hundreds of documents.

Engineers and designers need a faster, more accurate way of verifying requirements, which is where artificial intelligence (AI) can play a vital role. To support this Mott MacDonald, has developed a bespoke AI tool known as REALM – Requirements Evaluation and Analysis for Lifecycle Management.

The tool has been trained to analyse vast amounts of text and elicit requirements much more efficiently and reliably than can be done manually. It can help engineers to verify project deliverables, clarify client expectations and ensure a consistent interpretation of guidance and standards. Importantly, our people remain in the loop throughout and every output requires human review, interpretation and judgement.

REALM removes monotonous manual work and creates more time for our engineers to apply their judgement to problems, which is central to our responsible use of AI. The tool has already yielded substantial time and cost benefits on major projects when integrated into our already well-established engineering processes. As it becomes an integral part of our approach to delivery, it has huge potential to help engineers bring greater precision and value to projects, identify potential discrepancies and scale up delivery capacity.

Three valuable dimensions of REALM

Module 1: Elicitation

This AI-powered tool identifies key phrases such as "shall" and "must" in core project documents and extracts a comprehensive list of client requirements for review by a specialist team. On its pilot project, it was 35% to 45% faster at compiling 4,000 potential requirements from 42 documents than manual processing. After extensive testing and validation, this module   is being used on projects and has delivered efficiencies of up to 80%.

Module 2: Quality analysis

This module uses AI to identify incomplete and contradictory requirements. By classifying requirements as red, amber or green, it flags issues that need further investigation by the requirements team and creates a more targeted review process. We are currently testing the performance of this module at picking up contradictory requirements.

Module 3: Validation

This module uses AI to simplify the process of providing continuous assurance updates on project deliverables. It helps the systems engineering and assurance team to carry our periodic compliance checks in less time with minimal disruption to core engineering activities. This module is being shaped to meet the needs of the business through discussions between key stakeholders.

REALM uses rules-based logic to extract, verify and categorise technical requirements from complex documents

Keeping the human in the loop

The core team who developed REALM have a systems engineering and assurance background with extensive experience on major infrastructure projects. Working with AI experts in the digital team, they began developing the tool in June 2024. Over the first six months, they developed the script and logic for module one. Through systematic testing, the team fine-tuned the AI tool to a point where it can identify mandatory requirements and highlight grey areas that require professional judgement.

REALM is a static model, which means it doesn’t self-learn or autonomously adapt. Its value is that it can complete repetitive, low-value tasks with accuracy. Humans remain in the loop throughout and the tool has been designed to meet all of our AI governance requirements.

Promoting common understanding

For engineers, establishing project specifications is a vital step before the design of an asset begins. It is the basis for common understanding between stakeholders Ambiguity in product documents can either lead to extra unnecessary work being done or less work being done than is required.

For example, there might be a situation where a contract states that an engineering team is responsible for designing lighting systems, but the scope of works implies that they are also responsible for the background power supply that feeds the lighting.

Module two of REALM can help to identify this kind of discrepancy early on, triggering valuable client conversations about whether to remove an existing requirement, clarify the wording or add a new requirement.

The module, which is still under development, can provide the evidence base to support project modifications via a change variation request. This could lead to additional revenue for carrying out a wider scope of work, and potentially higher profits.

On ambitious projects, engineers need time to analyse project risks and ensure a consistent interpretation of standards and guidelines. Using AI for onerous, repetitive tasks, helps engineers to establish more quickly where the gaps or risks lie on a project.

If certain requirements are unrealistic, engineers can apply their expertise and judgement to find a constructive, pragmatic way forward. Bringing clarity to discussions with clients and partners leads to stronger collaboration, fewer disputes and better project outcomes.

Moving to a continuous cycle of assurance

Assurance is still often thought of as the end point of project delivery, where engineering firms demonstrate what they have delivered. But looking to the future, AI tools can support the shift towards ‘progressive’ assurance of projects throughout a project’s lifecycle, rather than delaying validation of deliverables until the end.

Where module three of REALM adds value is by providing a clear audit trail of what happened when. The tool can extract relevant evidence on the ongoing status of each deliverable – what is the requirement? What are the acceptance criteria that need to be met? How does it relate to the original contract? Again, while keeping humans in the loop, AI processing power has the potential to improve the speed, accuracy and consistency of compliance checks and client reporting.

So from beginning to end, REALM has the potential to deliver benefits across the project lifecycle. With module one already empowering people to generate value for clients, we are continuing to shape AI to develop future modules that improve quality, ensure a common understanding and enable deeper collaboration, with major cost saving potential.