Mott MacDonald: Engineering a responsible agentic AI future
Through its AI transformation journey, the engineering consultancy is learning how to improve project delivery and strengthen decision-making for clients, while protecting data, boosting quality, and preserving human judgement.
Mott MacDonald’s AI journey with Microsoft Copilot began with just 50 licences. Today the business is scaling AI across its 20,000 people through a major investment in Microsoft 365 E7 – Microsoft’s top-tier subscription encompassing work apps, AI assistant, and enterprise security in one package. And colleagues have now created more than 8,000 AI agents.
Those figures show how quickly things have progressed at the global engineering consultancy since we last spoke to Mott MacDonald. The focus then was on its AI assistant EMMA and employees beginning to build their own AI tools.
Copilot has spread much further across the business since, and its people are using AI on live projects, developing specialist agents to support everything from project assurance and stakeholder engagement, to design automation and data-driven decision making, says Nasrine Tomasi, Global Head of AI and Data at Mott MacDonald.
The result is an emerging ecosystem of thousands of agents, helping the company automate complex workflows, augment expertise, and scale innovation responsibly across disciplines and geographies.
“A lot has happened in the last nine months!” says Tomasi.
Taming complexity
AI is being used to work through the dense technical documentation behind major infrastructure projects. Contracts and designs can involve thousands of pages of requirements. Sometimes the same details appear several times, in different documents, and sometimes those requirements contradict one another.
On one airport project, for example, AI helped Mott MacDonald work through around 220,000 requirements and flag about 200 conflicts for attention.
Finding these discrepancies early can prevent redesign, delay, and extra cost later.
On another project, involving a large underground station, structural designers used Copilot-supported analysis to show that temporary works could safely form part of the permanent structure, reducing the amount of additional construction required.
Bidding for new work is changing, too. A hydropower team has built a collection of AI agents to help prepare business proposals for new projects. They assemble requirements, calculate budgets, and find matches to the company’s capabilities.
Team members refine the agents after each bid and have set themselves a target of cutting the average proposal process from one week to 48 hours.
“They are all contributing to the agents to make them better with every bid,” says Tomasi. “And they are winning more work as a result.
“We had talked a lot about productivity and doing things faster, especially a year ago,” Tomasi says. “But we are now going way beyond this, driving up quality and creating value for our clients.”
‘Control tower’ for agents
Success creates its own challenges, though.
Among Mott MacDonald’s 8,000-plus agents are innovative ideas, specialist project tools, and solutions that could be scaled across the business. But rapid growth has led to some overlapping solutions, dormant agents, and knowledge gaps when creators leave or change roles.
To tackle this, the company is creating a central AI library, so if someone is looking to develop an agent for proposal writing, for example, they’ll be able to discover whether one already exists that they can either use or adapt.
The aim is not only to improve visibility, but also to identify successful innovations with the potential to scale across the organisation.
As the agent ecosystem grows, the business needs a clear understanding of who owns each agent, what data it can access, how it is being used, and whether it continues to deliver value, explains Tomasi.
Microsoft Agent 365 will provide that oversight, giving Mott MacDonald a single view of its growing agent ecosystem and the controls needed to manage it responsibly. Tomasi describes it as a future “control tower” for agentic AI, helping the business balance rapid innovation with enterprise-grade governance.
“Protecting client information is crucial,” she says, “as is maintaining the quality of what we do as a business.”
New creativity
Of course, this kind of transformational change didn’t happen overnight. The roll-out has been gradual. After obtaining its first 50 Copilot licences following its introduction in 2025, Mott MacDonald expanded to 1,000, then 5,000. Half of the new licences in the second stage went to early-career professionals, many of whom were already comfortable using generative AI.
That led to what Tomasi calls “a little bit of reverse mentoring”, with younger employees learning from experienced engineers while showing those same colleagues what AI could do. Many have since become AI champions within the business.
The experience changed Mott MacDonald’s thinking about access. Rather than trying to predict where AI would prove most useful, the company saw value in giving people the tools and letting them apply their own expertise.
“We’ve got very creative people. They love to think about how they can solve a problem,” Tomasi says. “Giving them access to these tools is just creating another layer of creativity.”
That thinking now extends to the whole workforce, with Copilot rolling out to all 20,000 employees under a new five-year agreement.
EMMA’s new job
The company’s AI assistant EMMA – Every Mott MacDonald Answer – is now moving beyond simple knowledge retrieval and getting involved in business processes, says Tomasi.
Monthly project reviews are one example. Project managers and project principals previously had to complete a standard form covering the health and status of a project. Now, using an app hosted in Microsoft Teams, EMMA can capture a structured conversation between them and organise the notes around an agreed agenda, highlighting lessons learned and future actions. More than 10,000 lessons have been captured since the app launched in May.
“The richer the conversation, the richer the notes that you’re having at the end of this process,” says Tomasi.
Mott MacDonald is also preparing to use Microsoft Fabric as a central data platform, giving Copilot and its agents better access to the information they need.
Experts still decide
As AI becomes part of more embedded within Mott MacDonald, the company is also trying to make sure employees understand where it can go wrong.
Its adoption programme is intended to help people use the tools confidently while still critically reviewing their output. Technical leaders need to spot weak answers, understand the limits of the technology, and know when human judgement should take over.
For Tomasi, the current goal is to make that confidence widespread.
“It’s about creating a community that’s confident in its use of AI and can use it to keep shaping the future of Mott MacDonald and the infrastructure that people use every day.”