Sales CRM app prompt with pipeline and follow-ups

Create a sales CRM for [TEAM TYPE] that tracks [SALES PROCESS]. The first release serves [USER ROLES] and must answer: which deals need a next action today?

Before implementation, describe the smallest data model for organizations, contacts, deals, stages, activities, tasks, and users. Explain the relationships and which fields are required. Then build:
1. Pipeline view with stages [PIPELINE STAGES], deal counts and totals, owner filter, and an accessible alternative to drag-and-drop.
2. Contact and company records with search, duplicate detection, notes, and linked deals. Do not merge records silently.
3. Deal detail with value, stage, owner, expected close date, activity history, and one explicit next action.
4. Follow-up queue ordered by due date, including overdue, due today, completed, and no-next-action states.
5. A compact dashboard showing open pipeline, overdue follow-ups, and stage movement using internally consistent numbers.

Behavior and security:
- Every create, edit, stage move, task completion, and deletion must persist or be clearly labeled as demo behavior.
- Validate fields on both client and server; show useful errors, empty states, and loading states.
- Enforce role permissions for [USER ROLES] at the data layer. A user must not gain access by changing a URL or client-side filter. Include an audit trail for deal ownership and stage changes.
- Never put customer data or secrets in client logs. Use fictional seed data marked as sample.
- Export contacts and deals as CSV with predictable column names, excluding fields the current role cannot view.

Visual direction: [VISUAL DIRECTION]. Prioritize dense but readable tables, clear status, and fast keyboard navigation. Avoid a decorative dashboard that hides the follow-up queue.

Finish with a test plan using two accounts with different roles. Verify permissions, duplicate handling, stage movement, overdue tasks, export, and mobile access to the next-action queue. State any backend or auth integration still needed before real customer data is imported.