The AI lifecycle on the platform
Every AI asset follows the same path: use case in, build, independent validation, deploy in-region, monitor, reuse, and value tracked.
For: Everyone · 1 min read
The stages
| Stage | What happens | Who is accountable |
|---|---|---|
| 1. Use case | A business unit raises an idea. It is matched to existing assets first, then sized and prioritised. | Sponsoring BU, AI Centre of Excellence |
| 2. Build | Data scientists build on the lakehouse and Vertex AI, with datasets they are entitled to. | Builder |
| 3. Validate | Model Risk Management validates independently, in proportion to the model's risk tier. | Model Risk Management |
| 4. Deploy | The approved version is deployed to an endpoint in Google Cloud Malaysia. | MLOps |
| 5. Monitor | Drift, performance and fairness are checked against the thresholds set at validation. | MLOps, model owner |
| 6. Reuse | The asset is published in the catalogue; other business units subscribe and are charged back. | Marketplace operator |
| 7. Value | Realised value is tracked against the business case. | Sponsoring BU, Group Finance |
Why one path for everything
A single path means Internal Audit and the regulator can ask the same questions of every model and get an answer from the platform, not from a spreadsheet. It also means reuse is the default: before anything is built, the platform checks whether an asset already does the job.
Models rebuilt from SAS
Several catalogue models are marked as rebuilt from SAS. They went through the same lifecycle, with the old SAS model kept as the champion until the Python rebuild matched or beat it in parallel running.
Read next
Was this helpful?