This evening I babbled a thought stream to one of my colleagues about how generative AI (GenAI) is transforming many aspects of business, but its application in enterprise architecture presents a unique paradox. While it’s a powerful tool for accelerating and automating existing processes, its reliance on historical data and patterns may cause its outputs to ‘regress to the mean,’ preventing it from making truly transformative or innovative leaps. This doesn’t mean GenAI is useless in this domain; rather, its value lies in specific, targeted applications. The rest of this post elaborates on the thought.
The “Regression to the Mean” Phenomenon
The core of this argument lies in how GenAI models are trained. They learn from vast datasets of existing information, identifying patterns, relationships, and common practices. When applied to enterprise architecture, these models become experts at what has already been done. They can analyse a company’s current architecture, understand industry best practices, and even generate diagrams and documentation based on established frameworks.
For example, a GenAI model can be excellent at:
- Automating documentation: It can quickly create and update architectural diagrams, API documentation, and process maps by ingesting existing data.
- Roadmapping: By analyzing past project data and business goals, it can generate roadmaps that follow established patterns for technology adoption and change management.
- Compliance and risk assessment: It can check a new architecture against a set of compliance rules and historical risk data, flagging potential issues.
This capability is highly valuable for improving efficiency and ensuring consistency. However, since the output is based on a distillation of what already exists, it can struggle to create something genuinely novel. It’s like asking a model trained on a library of classical paintings to invent cubism; it will produce masterful imitations, but it’s unlikely to create a revolutionary new style.
The Limit of Innovation in Enterprise Architecture
Enterprise architecture isn’t just about documenting the present or optimising the past. It’s about envisioning a future state that may not exist yet. This requires abstract reasoning, strategic foresight, and a creative leap of imagination that goes beyond the patterns of a training dataset. It involves asking questions like:
- How can we completely redefine our business model using a technology that’s just emerging?
- What new capabilities can we build that will fundamentally change our industry?
- How can we design a system that addresses a problem no one has ever solved before?
These are areas where GenAI’s “stochastic parrot” nature—its ability to mimic without true understanding—becomes a significant limitation. While it can suggest solutions based on what’s worked before, it’s not equipped to make the kind of counter-intuitive or visionary jump that a human architect can.
The Power of Cross-Pollination
The true transformative power of GenAI in enterprise architecture isn’t in creating novel architectures from scratch, but in facilitating cross-pollination of knowledge between different industries.
Enterprise architecture often struggles with siloed thinking. Architects in one industry might not be aware of common solutions or architectural patterns used in another. GenAI, with its vast and varied training data, can act as a bridge between these domains. For example:
- A financial services company can use GenAI to explore how a manufacturing company’s supply chain optimisation architecture could be adapted to improve its own data flow and business processes.
- A healthcare provider could draw inspiration from the distributed ledger technologies used in logistics to create a new, secure system for patient data management.
By using GenAI as a tool to rapidly surface and use architectural knowledge from diverse sources, we can apply proven solutions to new problems, sparking innovation through analogy and adaptation. It turns the model’s regression to the mean into a strength, as it efficiently brings the “mean” of one industry’s best practices to another’s unique challenges.
