In our published study, where a page sat in the site graph predicted its business role better than reading the page — 95% on that estate. That is why Digital MRI maps topology before content.
Author: Imre Lóránt Dévai
Paper: Building an AI Governance System for Enterprise Level Websites
Results from one large enterprise estate. Numbers are study outcomes, not a universal guarantee — the method is what we apply elsewhere.
Topology as the primary signal
On the evaluated site, network position alone predicted page business function at 95%. Degree, centrality, and cluster membership revealed organizational role.
Structure was the dominant signal, not content
Network vs. text features
Network features outperformed text-based NLP by 34.3 percentage points on that estate. Content captures what a page says; topology reveals what it does.
Position in the graph mattered more than keywords
Math matched the org chart
Detected communities had 94.2% homogeneity with actual business units on that estate. The mathematical structure mirrored how the organization was organized.
Network communities reflected real business divisions
Can we tell what a page does in an organization from where it sits in the link graph — without reading a word of content?
Enterprise sites are networks, not document piles. A page’s place in the link graph should predict its business role better than analyzing its text.
Keywords and topics describe what a page says. They often miss how pages connect — the relationships that define purpose in a large estate.
Fresh pages start with no network position. For those, text-based classification still works well until the graph catches up.
We analyzed a large enterprise website: full link graph, network features for every page, then classifiers that predict business function from position alone.
Crawl every page and map internal links into a complete site graph.
Compute topological metrics such as degree, PageRank, centrality, clustering, and community membership.
Detect communities in the graph — groups of pages that link densely to each other.
Train classifiers to predict each page’s business role from network position alone.
Key insight: Pages that link together tend to serve similar business purposes. The network structure encodes organizational logic.
Discovering nodes...
Practical implications for enterprise web governance and AI readiness.
Page-level SEO and content audits miss the bigger picture. How pages connect, cluster, and flow is the primary signal for understanding the estate.
For governance, why matters as much as what. Transparent models support audit trails and stakeholder trust that black-box AI cannot.
Deep learning added cost and opacity without an accuracy advantage on this task. Use the simplest approach that works.
Content without network context can still be classified with text features until links form. Established and new pages both get enterprise-grade coverage.
The same network-science lens powers Digital MRI: map the public site graph, measure reachability, and show what crawlers and AI agents miss.
Anonymized Fortune 500 topologies under the Five-Lens framework: patterns and failure modes from real enterprise sites.
The founders, the research, and a Budapest practice applying network and data science to AI governance.
Tell us a little about the site or the problem. We reply within one business day.
