Every AI initiative we're asked to rescue has the same post-mortem. The model was fine. The demo was impressive. Then it met the ERP: three definitions of "customer," custom tables nobody can explain, and processes that exist in four regional variants for historical reasons. The AI didn't fail — the core did.
The core is the dataset
Enterprise AI is only as good as the operational data underneath it, and for most businesses the system of record for that data is SAP. Forecasting, intelligent automation, agents that act on purchase orders or billing — all of it reads from and writes to the core. A messy core doesn't just slow AI down; it quietly poisons every answer.
This is why we treat clean core as an AI strategy, not just an upgrade philosophy. A clean core means standard processes wherever the business allows, extensions moved out of the kernel and onto side-by-side platforms, governed master data, and documented APIs instead of point-to-point workarounds.
What it unlocks, concretely
With a clean S/4HANA core, three things become possible that are painful otherwise:
- Trustworthy signals. Forecasting and anomaly detection run on live, consistent data — not extracts that took three weeks to reconcile.
- Agents that can act. When processes run through standard, documented interfaces, an AI agent can safely create a purchase requisition or release a billing block — with permissions and audit trails the business already understands.
- Faster time to value. Each new use case reuses the same governed data and APIs. The second project costs half the first; the fifth is a configuration exercise.
Getting there without stopping the business
You don't need a finished transformation to start. We advise clients to treat the S/4HANA move as a data programme wearing a migration's clothes: measure custom-code debt honestly, retire what nothing uses, standardize the processes with the highest AI upside first, and route every new extension through the side-by-side platform from day one.
The order matters. Enterprises that clean as they migrate arrive with a core that's ready for what's next. Those that lift-and-shift arrive with the same problems on newer hardware.
The takeaway
- AI value is bounded by ERP data quality — fix the core before scaling the models.
- Clean core = standard processes, side-by-side extensions, governed master data, documented APIs.
- Sequence your migration around the processes AI will touch first.
- Every use case after the first should get cheaper. If it doesn't, the core isn't clean yet.
Our SAP and AI practices work as one team on exactly this seam — see SAP services and AI services, or talk to us about where your core stands.
