How Customer City Works
A graph-based approach to customer data quality—built for RevOps teams, designed with engineering rigor.
System Overview
Customer City operates as a control plane for customer data quality. We connect to your systems of record, build a unified graph representation, compute quality metrics, and optionally write back approved changes.
Read-First by Default
All integrations start read-only. Write-back requires explicit configuration and approval workflows.
Your Data Stays Yours
We process metadata and compute derived metrics. Raw PII is encrypted at rest and never used for model training.
Stateless Compute
Scoring and rule evaluation are deterministic and reproducible. Same inputs, same outputs.
Graph-Based Customer Representation
We model your GTM data as a property graph where entities are nodes and relationships are edges. This structure enables cross-object quality rules that table-based tools can't express.
Node Types
Account— Company or organization entityContact— Individual person with contact infoLead— Unqualified prospect, pre-conversionOpportunity— Revenue opportunity with stage/valueActivity— Emails, meetings, calls, tasks
Edge Types
HAS_CONTACT— Account → Contact relationshipHAS_OPPORTUNITY— Account → OpportunityCONTACT_ROLE— Contact → Opportunity (with role)ACTIVITY_ON— Activity → Contact/Account/OppCONVERTED_TO— Lead → Contact + Account
Properties
Each node and edge carries properties from source systems plus computed metadata: confidence_score, last_verified, anomaly_flags, source_system.
Confidence Score Computation
Every record receives a Confidence Score (1-100) based on rule evaluation. Scores are explainable, auditable, and tied directly to rule evidence.
Rule-Based Foundation
Scores derive from explicit rules, not opaque ML models. Each rule has defined criteria, pass/fail logic, and assigned weight.
Weighted Aggregation
Rules carry configurable weights reflecting business priority. Teams can tune weights without changing rule logic.
Cross-Object Evaluation
Rules can span relationships: "Closed-Won Opp must have Champion contact role" evaluates across Opportunity and Contact nodes.
Evidence Trail
Every score includes evidence: which rules passed/failed, contributing factors, and lineage to source data.
Scoring Model
Confidence = Σ(rule_result × rule_weight) / Σ(rule_weight) × 100 Where rule_result is 1 (pass) or 0 (fail), and weights are positive integers configured per rule.
Quality Rule Framework
Rules are the building blocks of data quality. Our rule engine supports single-object validation, cross-object constraints, and temporal conditions.
Single-Object Rules
Validate properties on individual nodes.
Account.Industry IS NOT NULL Contact.Email MATCHES email_pattern Opportunity.Amount > 0 Cross-Object Rules
Validate relationships and constraints across nodes.
Opportunity HAS ContactRole WHERE role = 'Champion' Account HAS AT LEAST 1 Contact Contact.AccountId REFERENCES valid Account Temporal Rules
Validate recency, staleness, and time-based conditions.
Account HAS Activity WITHIN 90 days Opportunity.DaysInStage < 60 Contact.LastModified WITHIN 180 days Connector Architecture
Secure, configurable connectors for your GTM systems. OAuth-based authentication, granular permissions, and optional write-back with approval controls.
Salesforce
HubSpot
Email & Calendar
Enrichment
Grounded AI Assistance
AI capabilities that augment human decision-making—never replace it. All recommendations are explainable and tied to graph context.
Rule Generation
Describe intent in natural language; AI proposes rule structure with appropriate node/edge references and suggested weights.
Score Explanation
Ask "why is this score low?" and receive a natural language summary of failed rules, contributing anomalies, and suggested remediation.
Anomaly Triage
AI groups related anomalies by likely root cause, suggests ownership routing, and prioritizes by downstream revenue impact.
Value Recommendation
For missing or invalid fields, AI suggests best-fit values from enrichment sources or cross-referenced graph data.
AI Design Principles
Grounded in Data
AI operates on your customer graph, not generic training data.
Explainable Outputs
Every recommendation cites evidence from rules or graph context.
Human in the Loop
AI proposes; humans approve. No autonomous data modification.
Auditable Decisions
All AI interactions logged with inputs, outputs, and user actions.
Enterprise Security Posture
Infrastructure
- Hosted on AWS (US regions, EU available)
- VPC isolation per tenant
- Encryption at rest (AES-256) and in transit (TLS 1.3)
- Daily backups with point-in-time recovery
Access Control
- SAML 2.0 SSO (Okta, Azure AD, Google)
- SCIM provisioning support
- Role-based access per Graph Space
- API keys with scoped permissions
Compliance
- SOC 2 Type II certified
- GDPR data processing compliant
- Annual penetration testing
- Vendor security assessments available
Audit & Monitoring
- Complete audit log of all actions
- Change history with user attribution
- Real-time anomaly detection on access patterns
- Log export to customer SIEM (optional)
Ready to see it in action?
Schedule a technical walkthrough with our team. We'll show you the graph model, rule engine, and scoring in your context.