The MIT NANDA study found that 95% of enterprise AI investment — between $30Bn and $40Bn — generated zero measurable return. The consistent failure mode is bolt-on: AI deployed as a feature on an unchanged platform, without a governed knowledge substrate. The practice's Systematic AI architecture is the structural response to this failure mode.
What Systematic AI Means
Systematic AI is an architectural category — not a brand name or a product. It describes the governed convergence of four components where each governs and amplifies the others:
Knowledge Graph — structured intelligence. The organisation's own data, relationships, and domain knowledge encoded in a queryable, governed graph.
GraphRAG — retrieval against that structure. Not retrieval from unstructured documents. Retrieval from a governed substrate where the quality of output is bounded by the quality of the graph.
Agentic AI — autonomous task execution. Agents that operate within the governed structure rather than free-ranging across unstructured data.
LLM — language reasoning. The conversational and analytical layer that makes the structured intelligence accessible and actionable.
Each component alone has a well-documented failure mode. Together, in governed convergence, they produce AI that learns from the organisation's own data, compounds its intelligence over time, and creates switching costs no workflow competitor can replicate.
In Production — Not In Theory
The practice has two Systematic AI systems in production. PRAXIS — the practice's programme intelligence platform — runs on a Neo4j knowledge graph with 99 automated governance constraints and an AI-powered advisory layer. It governs every engagement the practice delivers. The methodology is running software.
The WN Layer
In the Architectural Formula, WN represents the compounding intelligence layer — what happens when the platform substrate learns. Businesses operating only at the workflow level (X) are fully disruptable by agentic AI. Businesses operating across workflow and hierarchy (X × Y) remain defensible. The intelligence layer (Z) creates the compounding moat. WN is where AI transforms what the software fundamentally is — not what it does.