Enterprise Architecture🇬🇧 English
AI in Enterprise Architecture: Governance
July 2025·7 min read·Brian P.N. Tofft
Artificial intelligence is not simply another IT system to implement and operate. It represents a fundamental shift in how organisations collect, process and apply information, and it places entirely new demands on Enterprise Architecture. Organisations that fail to account for AI in their EA framework risk ending up with a fragmented AI portfolio, redundant data silos, and strategic investments that do not cohere. At We Lead Projects, we have helped organisations navigate exactly this challenge. Here are our insights.
Traditional EA frameworks such as TOGAF are structured around four architecture layers: business, data, applications, and technology. AI touches all four simultaneously. At the business layer, AI changes decision-making processes and enables new business models. At the data layer, AI places heightened demands on data quality, data lineage, and access management. Poor data produces poor AI outcomes. At the application layer, AI introduces new components such as models, inference endpoints, and MLOps pipelines. At the technology layer, AI workloads often require specialised infrastructure including GPU capacity, cloud platforms, and data platforms. An Enterprise Architect who has not mapped these interdependencies creates blind spots that lead to misguided investments.
AI governance is the component of EA that ensures AI systems are developed and deployed in a responsible, reproducible, and controllable manner. It encompasses: model registration and version control, data rights and compliance (including the EU AI Act and GDPR), fairness and bias evaluation, auditability and explainability, and clear ownership structures for AI assets. Many organisations today have AI experiments scattered across business units, in marketing, finance, operations, and HR, without central coordination or a governance policy. An Enterprise Architect who establishes an AI governance framework brings these initiatives under a common structure, reduces risk, and increases the reusability of data and models across the organisation.
One of the first and most important steps in AI-EA work is creating an inventory of existing AI assets: which models are in use, who owns them, what data were they trained on, and which systems are they integrated with? This mapping typically reveals three things: (1) Many AI solutions are purchased as part of SaaS products and are invisible in the existing IT portfolio. (2) There is significant redundancy. The same problem is being solved with different AI solutions in different parts of the organisation. (3) Data quality varies considerably across the data sources used for AI. This inventory forms the foundation for a realistic AI roadmap.
An EA-based AI roadmap is not a wish list of AI projects. It is a prioritised plan grounded in the organisation's strategic objectives, mapping the architectural prerequisites and identifying the critical capabilities, typically a data platform, governance structures, and competencies, that must be in place before major AI investments begin. The roadmap should distinguish between three types of AI initiatives: (1) Buy: purchasing AI functionality as part of existing software. (2) Customise: fine-tuning existing models for the organisation's specific needs. (3) Build: developing proprietary models on proprietary data. Most organisations need all three, but in different proportions depending on their data maturity and strategic ambitions.
At We Lead Projects, we have led and supported AI implementation projects across multiple phases, from initial needs analysis and architecture assessment through to delivery and transition to operations. Our experience is that projects which succeed share three characteristics: first, a clearly defined business problem to solve, not 'we want to do AI', but 'we want to reduce the processing time for X by Y per cent'. Second, a solid data foundation established before model development begins. Third, a governance structure defined and approved by leadership before the first model goes into production. Projects that fail typically start with the technology and work backwards towards business value. The wrong order.
Wherever your organisation is on its AI journey, an architecture assessment is a strong starting point. It clarifies what you have, what you lack, and what the right next step is, without committing you to a specific technology or vendor. At We Lead Projects, we offer AI-EA assessments as a scoped engagement, typically four to eight weeks, delivering a concrete report and a prioritised roadmap. Contact us to learn more.
Brian P.N. Tofft
Managing Partner, We Lead Projects
Brian has more than 30 years of experience in project management and IT transformations across industries.
Get in touch and find out how we can help with your next project.
Contact us