AI build guide
Build a CRM with AI around a real sales workflow
A useful CRM reflects the team’s process. Start with the decisions people make, then model contacts, organisations, opportunities, activity, and ownership around them.
01
Design the record model
Separate organisations, people, opportunities, stages, activities, and assignment history.
02
Make actions accountable
Enforce access on the server and record important changes so teams can understand who changed what.
03
Test the working loop
Verify create, update, search, filtering, stage changes, and permission boundaries with realistic data.
04
A production-shaped Xroga prompt
Build a sales CRM for two teams. Model companies, contacts, opportunities, stages, owners, notes, and an immutable activity trail. Enforce team access on the server, support import validation and duplicate detection, and test search, reassignment, stage changes, and unauthorized reads.
05
Evidence to require
Inspect role checks, database constraints, audit events, realistic seeded flows, and export behavior. A contact table alone is not a verified CRM workflow.
06
Before calling it production-ready
Review permissions, data handling, error states, accessibility, mobile behaviour, tests, environment configuration, observability, and rollback. The exact checklist must follow the project rather than a fixed framework template.