Picture a global FMCG manufacturer with thousands of ETL jobs quietly running overnight, some of them chewing through 18 to 20 hours just to merge ERP data with everything else the business needs to make a decision. Somewhere in that landscape sit 12 to 18 terabytes of BW code, built up over a decade or more, with nobody entirely sure anymore which reports still matter and which are relics nobody dares switch off.
Multiply that story across an industrial supplier remapping its finance data for S/4HANA, and a power utility trying to predict transformer failures from sensor data that has nowhere obvious to live, and you have a fairly accurate snapshot of enterprise data today: powerful, valuable, and quietly buckling under its own weight.
These aren’t hypothetical scenarios. They’re real customer journeys Mayank Madan, Head of the Data Practice at Lemongrass, works through every week. In the latest episode of Lemongrass Roots, Mayank sits down with Eamonn O’Neill, CTO and Co-Founder of Lemongrass, to unpack a topic customers are asking about more than almost any other right now: SAP Business Data Cloud (BDC), and why it has become a critical foundation for SAP’s new Business AI Platform (BAIP) message.
With 22 years in the industry, including time at Hadoop pioneer Hortonworks, Mayank has lived through several distinct eras of data architecture firsthand. This conversation traces that arc, from the classic data warehouse to today’s data fabric, and turns the stories above into a practical playbook for making enterprise data ready for AI that understands business context, not just prompts.
A Short History of “Where Your Data Lives”
Mayank frames the evolution in two intertwined threads: the infrastructure behind data ecosystems, and the parallel rise of AI.
Twenty years ago, storage and compute lived together in classic data warehouses, and insight came slowly. Some ETL jobs ran for 18 to 20 hours just to merge ERP and non-ERP data into static dashboards. A decade or so later, the volume, velocity, and variety of data exploded, driven by mobile devices, sensors, and unstructured data. That gave rise to big data and, eventually, the lake house, with storage and compute finally separated, and technologies like Hadoop deployed at massive scale.
Now, says Mayank, the industry is moving again, from lake house to data fabric. Rather than another storage tweak, this shift is about building a semantic and knowledge layer that gives the business a single, trusted definition of its data, however people choose to consume it: natural language, dashboards, self-service analytics, embedded intelligent applications, or AI agents. That’s exactly where SAP’s BAIP narrative comes into focus: BDC provides the trusted, governed business data foundation; SAP Knowledge Graph adds the semantic context; Joule and agentic AI experiences can then reason and act on that context with greater accuracy and control. It’s also a more open ecosystem, with BDC designed to work alongside partners like Databricks, Snowflake, and Palantir rather than replace them.
For customers, that distinction matters. BAIP is not simply a new label for existing platform services. It is SAP’s attempt to connect data, process knowledge, application development, automation, and governance into one AI-ready operating model, so enterprises can build agents and workflows that are grounded in how the business actually runs.
IT and Business Are Finally Looking at the Same Problem – and AI Raises the Stakes
One of the most useful frames in the conversation is Mayank’s breakdown of how IT and business each experience data modernization differently, and why both lenses matter.
IT tends to worry about technical debt: years of warehouses, ETL tools, and point solutions that need consolidating. Business, meanwhile, cares about data silos and semantics – a single, trusted definition of a KPI like gross margin, so decisions can be made with confidence. In a BAIP world, that alignment becomes even more important because AI agents need the same governed definitions, policies, relationships, and process logic that people do. The shift in focus, Mayank explains, is from IT-built pipelines toward business-owned data products, where domain experts take ownership of data definition and quality while IT builds and governs the underlying pipeline. Success is measured differently too: IT tracks pipeline consolidation and deployment speed, while business wants real-time insight, trusted automation, and confidence that AI-driven recommendations are accurate, compliant, and explainable.
Eamonn draws a direct parallel to Lemongrass’s own experience giving project managers and delivery teams self-service access to their own data, where the very first question people ask is simply, “is this accurate?”
The Real-World Risks: Extraction, S/4 Remediation, and Legacy Consolidation
Talk to enough customers, and clear patterns of risk emerge. Mayank walks through three he sees constantly, and they’ll sound familiar from the opening.
Data extraction strategy. That FMCG manufacturer from the start of this article? It has thousands of custom ABAP extractors and ETL pipelines built up over 10 to 15 years to move data out of ECC and BW alone. Consolidating that sprawl compliantly and cost-effectively is now a central challenge. Mayank draws a clear distinction here: an extractor simply pulls data from one system to another, while a data product is a self-governed, domain-specific dataset – like a finance or inventory data product, that’s ready for business consumption out of the box, removing the need to build and maintain that extraction and transformation logic yourself.
S/4 remediation. This is the industrial supplier’s story: moving to S/4HANA changes data structures downstream (finance in ECC versus ACDOCA in S/4, for example), which can break years of custom extraction logic overnight. Mayank’s approach: map what data is actually needed, follow best practice extraction from CDS views, and lean on SAP’s managed data products plus BDC’s zero-copy capability, accessing data where it resides rather than replicating it, to cut both effort and risk. In the context of BAIP, this also helps create the clean, governed data layer AI agents need before they can safely support finance, operations, supply chain, or customer-facing decisions.
Legacy consolidation. And this is the power utility, trying to predict transformer failure from sensor data that has no natural home alongside its BW and SQL Server warehouses. For customers running multiple legacy platforms side by side, consolidation is a multi-year journey. Lemongrass’s answer is to blueprint the target ecosystem around actual business use cases first, then build the hybrid architecture, often combining BDC data products with tools like Databricks to support it. That use-case-first blueprint is increasingly important as enterprises move from analytics modernization to AI execution, where the value comes from connecting trusted data to real workflows.
Shrinking the Assessment Phase from Months to Weeks, and Building the AI Foundation
Perhaps the most striking number in the conversation: Mayank has seen 50 to 60% technical debt in legacy BW environments, with some customers carrying 12 to 18 terabytes of BW code built up over a decade or more. Understanding what’s actually being used down to the field level, is critical before any migration decision gets made.
This is where Lemongrass’s BW Insights accelerator comes in: an automated, end-to-end lineage tool that traces consumption (via Power BI, Tableau, or BEx queries) all the way back to the underlying data source, at both object and field level. The impact is significant, Mayank notes this can shrink an assessment phase that might otherwise take six to twelve months down to as little as eight weeks, while also feeding directly into S/4 impact assessment. It also gives customers a practical starting point for BAIP readiness by making clear which data should be governed, exposed as products, and trusted by people and AI alike.
Where to Start: People, Process, Technology
Asked for his single piece of advice to listeners starting this journey, Mayank returns to a familiar but often underweighted framework:
- People need cross-skilling skills, particularly teams coming from a BW background who need to adapt to new cloud-native tools and ways of working.
- Process has to evolve toward cloud-native deployment, CI/CD, integrated monitoring and orchestration, and the governance controls needed to operate AI agents safely at enterprise scale.
- Technology decisions should be driven by close collaboration with the business, to make sure the right data products are being built for the people, applications, and AI agents that will actually use them.
The Bottom Line
The shift from data warehouse to lake house was a story about cheaper storage and bigger scale. The shift to data fabric, BDC, and now BAIP is different: it’s about ownership, trust, context, and speed to action, with AI woven directly into the fabric of how data gets consumed and work gets done. As Mayank and Eamonn discuss, the challenges – extraction, remediation, legacy consolidation – are well understood, and the accelerators and methodology now exist to de-risk the journey significantly.
Listen to the full episode of Lemongrass Roots with Mayank Madan, Head of the Data Practice at Lemongrass, to hear the complete conversation on SAP Business Data Cloud and what it means for your data platform.
