AI Needs More Than Data.It Needs Context.

Getting the right data to the systems that need it has remained a challenge through successive waves of technology, from Big Data and Data Science to Machine Learning and AI.

Böhm highlighted three factors that influence whether AI can be used reliably: trust, consistent quality and context.

If a finance controller questions a figure because it doesn’t match their own spreadsheet, they are unlikely to trust an AI agent using that same information. Accuracy alone isn’t enough, either. Data must be suitable for its intended use. A five per cent variation might be acceptable in a forecast but inappropriate in a profit and loss statement or regulatory report.

Maintaining quality is an ongoing task. As business requirements change, data that was once fit for purpose may no longer be reliable. AI can amplify these problems when flawed data feeds into automated decisions and workflows.

SAP BW landscapes contain years of accumulated business knowledge. ABAP code, mappings, transformations and calculation views often encode how an organization calculates revenue, builds forecasts and manages controlling processes. The people who originally developed this logic may have left, while the business still depends on it.

Moving everything to a new platform without examining it first risks carrying the same complexity forward.

Böhm recommends starting with a landscape analysis and end-to-end lineage mapping to understand which objects are used and which dependencies need to be preserved. In his experience, around 50% of legacy objects can often be retired, allowing teams to focus on the business logic that still matters.

AI-based explainer agents can help document existing code, while other agents can assist with code translation and quality checks. However, translating individual pieces of ABAP code isn’t enough when the underlying logic depends on reusable objects, hidden calls or customer exits. The business context and dependencies must be understood, too.

Böhm described a customer project involving 32 SAP systems across BW and ERP environments. The company wanted to understand cash flows across the group but lacked a consistent group-wide view.

One Data translated the existing BW logic into a modern Lakehouse architecture and built a group-wide model. The migration took months rather than years, and the improvement in managed cash was visible in the company’s profit and loss account.

The project illustrates how migration can address a business problem while modernizing the underlying technology. It can also provide an opportunity to invest in data quality and governance as part of a transformation companies already need to undertake.

A successful migration doesn’t automatically make data easy to understand or use. Organizations still need to maintain quality, preserve context and establish clear responsibility.

Böhm highlighted four practical requirements:

A data product approach can bring these practices together. Clear ownership, understandable metadata and defined quality expectations make data easier to manage and reuse across teams and use cases.

Technology can automate checks and improve transparency, but people still need to take responsibility for the data they manage. Governance remains an ongoing practice, not a project that ends at go-live.