Driving technical excellence at Mott MacDonald by building an enterprise agent on Azure

In today’s rapidly evolving engineering sector, global leaders like Mott MacDonald are under increasing pressure to deliver sustainable, innovative projects while maintaining technical excellence and client trust. As an employee-owned consultancy with a footprint spanning more than 150 countries, Mott MacDonald recognized that unlocking institutional knowledge and best practices across its teams was critical to driving project outcomes and maintaining its reputation for the highest quality. Facing the challenge of disconnected expertise and time-consuming information searches, the company set out to build an AI-powered solution that would democratize knowledge, accelerate delivery, and strengthen its competitive edge in the construction industry.

“For us as a company, technical excellence is a principle that’s core to our values, which means making sure that we deliver high-quality outcomes for our clients while keeping a strong commitment to sustainability and social outcomes,” says Nasrine Tomasi, Head of AI and Information Management at Mott MacDonald. “We saw AI as a facilitator for that. We wanted to build an agentic solution that would democratize best practices and expertise across the business.”

To accelerate knowledge discovery, the Mott MacDonald team decided to develop its first enterprise-wide AI agent dubbed Every Mott MacDonald Answer, or EMMA. In doing so, the company needed to protect organizational data to maintain its competitive advantage and client trust. “It was really important for us to curate the user experience and data that our agent could access,” says Paul Skidmore, AI Solutions Lead at Mott MacDonald. “Our people need to trust the responses that they’re given so they can use the information confidently and responsibly. Achieving this requires strong, well-governed data foundations to build upon.”  

Building an enterprise-ready agent on the pillars of responsible AI 

As a cloud-first company, Mott MacDonald’s employees use Microsoft SharePoint and Microsoft OneDrive for daily tasks. Understanding and organizing project and enterprise information from the company’s existing business tools was key to building an agent that can smooth workflows and help users solve problems based on company policy and best practices.

By building EMMA on Microsoft Azure, the company can enforce user-delegated access through its existing data controls. Furthermore, Mott MacDonald, which serves clients around the world, must comply with the European Union’s General Data Protection Regulation and Artificial Intelligence Act as well as regional-specific data residency requirements. “By architecting EMMA on Microsoft, we could accelerate our time to market while leveraging the existing infrastructure underpinning our organization’s security model. We knew that our data was protected and governed appropriately and we could meet data governance requirements,” says Skidmore.

To mitigate risk, Mott MacDonald began with a proof of concept for EMMA, which its teams initially built using Azure OpenAI Assistants available in Azure AI Foundry and Microsoft Graph API. This provided the AI reasoning and orchestration layer, using foundational models in Azure AI Foundry to interpret user queries and calling tools. By implementing Microsoft Graph API into the agent, Mott MacDonald can support secure, user-delegated access for organizational data stored in SharePoint sites and personal information stored in its employees’ OneDrive accounts.

While designing EMMA, Mott MacDonald prioritized security and governance from the beginning. The company conducted a comprehensive risk assessment, evaluating the entire AI system—including its architecture, privacy, data security, and ethical considerations—helping its AI team identify, assess, and mitigate potential risks. Additionally, a series of beta testing activities were conducted with a diverse group of users, whose ongoing feedback helped refine the agent’s responses and validate governance controls in real-world conditions. To monitor performance and guide improvements, analytics were made and stored in Azure Database for PostgreSQL, giving the AI team a clear picture of user interactions and helping uncover vulnerabilities.

To ensure model robustness before deployment, Mott MacDonald used the AI Red Teaming Agent in Azure AI Foundry. This process simulated potential attacks and misuse, giving the team the ability to evaluate how EMMA would behave in these scenarios. Additionally, the company uses Azure AI Content Safety as an ongoing protection layer to continuously block unsafe content. “EMMA is at the forefront of how we develop AI software responsibly at Mott MacDonald,” says Skidmore. “For example, we use AI Red Teaming Agent to help us perform robust evaluations of the models powering our solutions. We are taking key learnings from the process and disseminating them across the entire organization.” 

Expanding EMMA’s knowledge base using Azure AI Search 

The company’s proof of concept for EMMA evolved from using Azure OpenAI Assistants to Azure OpenAI Responses API, which brings together the best capabilities from the chat completions and assistants API in one unified experience. “Using the Responses API in Azure AI Foundry has helped us move faster without having to reinvent the wheel. File uploads and conversation histories are already covered, and we can focus on delivering intelligent answers for our staff,” says Evan Harwin, Data Scientist at Mott MacDonald.

After launching the first version of EMMA on the company’s internal intranet, the Mott MacDonald team enhanced search performance by integrating Azure AI Search into its architecture. This approach uses vector embeddings applied to curated internal knowledge, empowering EMMA to deliver more relevant results.

Combined with Microsoft Graph API, EMMA can provide secure, permission-aware access for user-specific data while delivering context-aware search capabilities across organization-wide knowledge. “Our ongoing investment in EMMA has been guided by data and insights from our user base,” says Skidmore. “We understand the topics people are asking about, the information they’re trying to find, and whether EMMA is doing well at handling their queries. We use those insights to drive positive change in Mott MacDonald around our infrastructure, data availability, and data quality.”

The integration of Azure AI Search marked a major step in broadening EMMA’s knowledge base. The company incorporated its technical knowledge and best-practices documentation covering more than 40 engineering and development topics, including aviation, geotechnics, and climate change, into the next version of EMMA. “The second major release of EMMA was a big step forward,” says Skidmore. “It’s when we really started to infuse the system with Mott MacDonald’s carefully curated knowledge, processes, and ways of working.” 

Democratizing institutional knowledge for 20,000 employees

Mott MacDonald rolled out EMMA to 20,000 employees in just nine months. Since then, EMMA has handled over 220,000 unique employee queries, reducing the time spent on information searching. The company has also accelerated onboarding and democratized best practices across the organization by reaching 60% of the business.

The project also strengthened data management practices and fostered a culture of information accessibility. “Our culture has evolved, and it’s been amazing to see,” says Tomasi. “Our experience developing EMMA is driving us further into our data journey. People want to have more organizational information accessible. It’s really shown the entire company the value of well-organized information and data.”

Looking ahead, Mott MacDonald will continue expanding EMMA’s knowledge base, with plans to integrate data from its customer relationship management and enterprise resource planning systems. As the company strengthens its agentic AI capabilities, EMMA is also evolving to become an AI partner for Mott MacDonald’s teams, supporting them in their daily workflows.

“We’re not only saving time on searching for information, but our teams are confident that they’re delivering the best to service our clients’ needs,” says Tomasi. “They can learn about innovations that we’ve delivered all around the world, which we can then bring to other projects. It’s really driving quality and technical excellence at Mott MacDonald.”

Discover more about Mott MacDonald on LinkedIn or Instagram.