Designing how enterprise AI understands relationships.

Company: ServiceNow
Role: Design lead
Collaboration: PM, Engg, QE, Design systems
Team: 8-person cross-functional team
Timeline: 5 months
Outcome: ServiceNow’s first knowledge graph builder
Patent: Co-inventor on patent filing
Launched: January 2025
Status: Live

I led the design of ServiceNow’s first knowledge graph builder, enabling administrators without graph expertise to define the relationships that give enterprise AI the context it needs.

Why connected context matters? Here's an HR use case showing how relationships between employees, skills, and training opportunities can give AI the context to make relevant recommendations.

The problem

Enterprise AI isn't limited by missing data. It's limited by missing relationships.

Every answer inside an enterprise already exists. The problem is that information lives across hundreds of disconnected systems that don't understand each other. A single prompt can require context from people, records, assets, and policies across separate systems.

The disconnect plays out across teams and systems. Different language, permissions, and reasoning models make it difficult for AI to connect enterprise knowledge accurately.

A network of connected questions illustrating the relationships enterprise AI must understand. — Each circle represents a data table. Even a seemingly simple task can require information from multiple tables working together. **The first design challenge was making these relationships visible to administrators who may not know which tables and connections are needed to complete the task.**

The design challenge

Humans naturally think in relationships. Enterprise systems think in records, tables, and workflows.

My job was to design the missing layer between them → a system that made enterprise relationships visible, understandable, and configurable for AI admins who had never touched graph theory.

Comparison of how people understand connected entities and how systems store them across separate data tables.

The Knowledge Graph Builder is, at its core, a context authoring tool. It doesn't just visualize data. It gives AI the relational context it needs to reason across systems.

Enterprise data sources connected through an authored relationship model to provide grounded context to AI.

Key concepts

The next section uses a few graph-specific terms. Use this quick reference before continuing.

AI admin

An admin is a ServiceNow persona who deploys and manages AI solutions across the organization.

Knowledge graph

A connected view of enterprise entities and relationships that gives data context and meaning.

Node

An entity represented by a ServiceNow table, such as Employee or Incident.

Edge

A relationship connecting two nodes, such as an employee “reports to” a manager.

Table

A list of records organized into columns, used as the data source for a node.

Graph Builder

A no-code workspace for creating graph schemas and configuring nodes, properties, and relationships.

Design decisions

Building a knowledge graph builder wasn't just designing an interface. Every major decision balanced graph theory, enterprise workflows, engineering constraints, and user trust.

Four questions shaped the product: where configuration happens, what types of edges exist, how relationships are defined, and how the graph stays usable at scale.

Technical concepts had to be clear and usable for administrators who were not data experts.

Where does configuration happen?

ServiceNow's existing tools all use panel-based configuration, but graphs are spatial. We chose the panel to reduce learning curve, then pushed the canvas to be an active trigger point.

Configuration interface showing where administrators set up a knowledge graph.

What types of edges exist?

Multiple edges between two nodes quickly collapse into noise. We grouped relationship types and surfaced identity at scale while keeping distinct tooltips.

Knowledge graph interface showing the available types of edges between nodes.

How should an admin define a relationship?

The team's mental model used tables, lists, and forms. Testing revealed that admins wanted to propose connections from existing schema, not begin from a blank canvas.

Interface for an administrator to define a relationship between enterprise entities.

How does the graph stay usable at 1,000+ nodes?

Enterprise knowledge graphs grow to thousands of nodes. Progressive disclosure, zoom behavior, and a lightweight full-graph state keep the experience understandable.

Large knowledge graph interface designed to remain usable with more than one thousand nodes.

Impact

Built from 0→1. Designed and shipped ServiceNow's Knowledge Graph Builder, making it possible for administrators without graph expertise to model and configure enterprise relationships.

Adopted across 8+ workflow areas. Expanded across HRSD, CSM, WSD, ITAM, ITOM, SecOps, Industry workflows and Finance, moving Knowledge Graph beyond its initial use case into shared platform infrastructure.

Became a foundation for AI experiences. The knowledge layer expanded into AI Search, AI Agents, Now Assist, NLQ and personalization, giving these experiences shared context about enterprise data and relationships.

Created a reusable platform capability. The builder became a shared component that could support domain-specific knowledge graphs rather than requiring teams to build graph tooling independently.

Patent pending. The interaction and system approach contributed to an ongoing patent application.