IDEAS AI platformIDEAS AI platform

Problems faced

Seven departments, seven wish lists, no baseline to compare against. Midway, leadership brought a target diagram far wider than the original design, and the per-head pricing model no longer reflected the value delivered.

How it was solved

1. Ran a current-state discovery and reduced conflicting requirements to a prioritised MVP: AI agent management, chatbot, a permissioned knowledge base and LMS integration.
2. Compared the target diagram with the baseline: of 49 blocks, 15 existed, 6 were partial, 6 were estimated and 22 were new.
3. Chose to extend the existing platform instead of a rewrite, under six "extend, don’t rewrite" rules.
4. Proposed a per-project commercial model: seven delivery packages (G0 to G5 and GR), each with acceptance criteria, paid on accepted milestones.
5. Re-planned enterprise SSO after finding the old proposal had not been started and did not match the stack: reviewed 254 endpoints and re-estimated with PERT times a risk factor at about 290 man-days, double the original five-sprint plan.
6. Set a go/no-go gate for the Lead Agent proof of concept on the ERP: found the real target CRM differed from the approved requirements, logged 5 scope deviations and agreed a four-week route with a scope-cut order settled in advance.
7. Ran DevOps: GitHub Actions CI/CD with security scans, rolling ECS deploys with automatic rollback.