Customer 360

A blueprint for cross-system customer intelligence: accounts, contacts, contracts, tickets, opportunities, invoices, and the conversations that connect them. Built for the case where a sales engineer or AE asks "what's open with Acme right now?" and the answer has to span the CRM, the ticketing system, and the billing system.

The graph

What this demonstrates

• Multi-source ingestion — five different sources unified by Account.Id as the canonical key.
• Account-centric navigation — every other entity has a one-hop path back to its Account.
• Graph-scoped chat — the AI assistant can be told to "stay on Acme" and the tool calls automatically scope to that account's subgraph.
• Cross-system aggregations — count open tickets, ARR by contract, days-since-last-meeting.

Retrieval

• Text search on Account.Name, Contact.Email, Opportunity.Name, Ticket.Subject — sales reps search by name.
• Hybrid search on Ticket.Body and meeting notes — for "what was the last conversation about renewal?"
• Graph-scoped: Account → Out(...) produces the target set for any account-scoped search.

AI

An Account assistant scoped to a single account per chat session. Tools:

• GetAccountSummary(accountId) — high-level: open tickets, active opportunities, ARR, days-since-last-touch.
• SearchInAccount(accountId, query) — hybrid search within the account's subgraph.
• RecentActivity(accountId, days) — events across all systems sorted by time.

Prompt template: "You are the Account assistant for . Answer using only the cited results from tools. If the user asks about a different account, ask them to switch context."

Permissions

Sales teams typically have read access to their assigned accounts only. Implement this by:

• Mapping CRM account ownership to a Workspace _AccessGroup per region or per AE.
• Restricting Account nodes with RestrictAccessToTeam at ingestion time.
• All children inherit ACL through OwnedBy ownership edges from their Account.

Customer Success and Support teams get broader read access via additional _MemberOf edges.

Connector

Run as one connector per source:

• Salesforce connector — Account, Contact, Opportunity. Cursor on LastModifiedDate.
• Ticketing connector — Ticket. Cursor on updated_at.
• CLM connector — Contract. Often slow-moving; daily full refresh is fine.
• Billing connector — Invoice. Daily full refresh.

Each connector is independent, runs on its own schedule, and identifies the canonical Account by a known cross-system mapping (commonly the CRM ID stored as a foreign key in the other systems). Reconcile orphans nightly.

Deployment

• Mid-size B2B has hundreds of accounts, thousands of contacts, tens of thousands of tickets — comfortably handled by a single 16 GB workspace.
• Connector failures should alert the account team, not just the platform team — surface them in Slack via a custom endpoint that posts a daily digest.
• Cross-system identity drift is the #1 operational pain. Schedule a reconciliation task and surface mismatches.

Common pitfalls in this domain

• Duplicate accounts from cross-system imports — the most common ingestion bug. Pin the canonical source (usually CRM) and resolve foreign IDs against it.
• Stale data feeling fresh — display a "last synced" timestamp per source so users know when the data was current.
• Over-broad permissions — sales reps shouldn't see all accounts by default. Audit _MemberOf regularly.

Related

• Connectors and Ingestion pipelines for the multi-source ingestion patterns.
• Access Control Model for ReBAC against multi-team ownership.
• Technical support — narrower scope, similar shape.