The promise of SAP HANA's in-memory engine for analytics has always outrun the practice. Most clients have a HANA Cloud or HANA on-prem instance, an ML team somewhere in the business with a model in a notebook, and no path between them. Closing that gap is more of an operations problem than a modelling one.
Where SAP and data science actually meet
- Predictive scoring inside HANA - PAL (Predictive Analysis Library) and APL (Automated Predictive Library) for models that need to run next to the data.
- BTP AI Core for models that are trained off-platform but deployed centrally with versioning and serving infrastructure.
- Joule / SAP AI services for the language-model layer over SAP data.
The operational story
A scored field is only useful if someone trusts it. Trust comes from being able to answer three questions on demand: which model produced this score, what data was it trained on, and when does it need to be re-trained. Building those questions into the pipeline from day one is cheaper than retrofitting them after the audit.