Cleanfield: How AI is reshaping the SAP world

The gap between where most SAP landscapes are today and where the platform is heading is greater than most managers realize.

I have been closely connected to this world for over three decades. As co-inventor of Hana and leader of the original Hana project and team, I helped build the foundation on which S/4 Hana runs today. As the CTO of Accenture's SAP Business Group for many years, I have led and supported numerous S/4 programs to successful implementation. In all these years, I have repeatedly encountered the same structural challenge - in different forms, but always the same at its core. 

How quickly companies can modernize and further develop their SAP landscape rarely depends on the platform itself. It depends on the custom code that has accumulated in SAP systems over decades - differentiating business logic whose documentation has never kept pace with the changes, and a legacy that has become both indispensable and increasingly difficult to master.

To solve this problem, I co-founded Nova Intelligence with Emma Qian and Sam Yang, who bring their cutting-edge AI expertise from Google DeepMind and Meta AI - joined on the leadership team by Justin Kershaw, the former CIO of Cargill, one of the largest SAP landscapes in the world.

AI reaches the necessary level of maturity

The decisive breakthrough lies in the interaction of this SAP architecture knowledge with modern AI. In the past 18 to 24 months, large language models and agentic architectures have reached a level of maturity that makes it possible for the first time to understand large, heterogeneous enterprise code bases not only syntactically, but also semantically and structurally. We have built Nova Intelligence on this basis: a platform that makes this technological progress productively usable in the SAP world for the first time - and uniquely in this form.

What has long been visible in the IT industry outside SAP has so far hardly been practically usable in the SAP environment. The reason for this is not a fundamental weakness of the AI models, but the fact that SAP is a proprietary, stateful system - with its own transport system, its own authorization logic, its own development tools and a high degree of context dependency. A generic AI co-pilot is not enough here. Only the combination of a deep understanding of SAP, modern agentic AI architecture and a platform that can work directly in this world makes the next step possible. The result is a platform that supports all SAP users - from Abap developers to functional consultants and business users - throughout the entire lifecycle: functional design, fit-to-standard analysis, modernization, development, documentation and daily productive operation. Nova is not a tool for a single task, but a permanent AI partner for the entire SAP landscape.

Deadline 2030

This challenge is most pressing in the upcoming transformation to S/4 Hana. With a view to the SAP migration deadline of 2030, companies are faced with decades of custom code that needs to be understood, cleaned up and, in many cases, rebuilt. This is precisely where the limitations of traditional approaches become most apparent. Brownfield, i.e. the technical conversion of the existing ECC system to S/4 Hana, is pragmatic, fast and attractive in the short term - but preserves a considerable amount of historically grown complexity and technical legacy. 

Greenfield, i.e. rebuilding S/4 Hana without taking over the legacy system, promises the cleaner target architecture, but in practice often proves to be lengthy, expensive and risky because the business-critical differentiation logic must first be fully understood and then rebuilt manually. Many companies therefore realize: brownfield starts faster, but remains expensive in the long term; greenfield is architecturally convincing, but is often difficult to manage economically and organizationally.

Intelligent cleanfield approach

Nova Intelligence is an agentic AI platform that was developed specifically for SAP. In view of the 2030 deadline, S/4 migration is currently the most urgent field of application - and this is where Cleanfield comes in: a migration method developed by Nova that is only made possible by Nova's agentic AI platform. Cleanfield combines the pragmatic advantages of a brownfield approach with the architectural quality of a greenfield target image: the existing landscape is neither simply technically migrated nor indiscriminately discarded, but systematically analyzed, compared against the SAP standard and only rebuilt in a Clean Core-compliant manner where real differentiation exists.

This approach is made possible by the technical architecture of the Nova platform. The first central building block is „AI Code Intelligence“. In mature SAP landscapes, the necessary transparency about custom code and its dependencies, usage, risks and business relevance is usually lacking. Nova analyzes this inventory not only statically, but also semantically and structurally. This results in a reliable view of the actual logic of the system: which programs are business-critical, which have only been updated historically, which risks are hidden in dependencies and data flows and where standard functions could already replace a significant part of the custom code today. A second central building block is the reconstruction of this logic in structured documentation at specification level. 

Fit-to-Standard

This makes implicit knowledge explicit and transfers it from the minds of a few experts into a reproducible knowledge base. This documentation forms the basis for well-founded „fit-to-standard“ decisions: What can be omitted? What can be transferred to the SAP standard? What is genuine differentiation and needs to be consciously rebuilt? This is precisely where the economic leverage lies. Every functionality that does not need to be continued permanently reduces maintenance, testing effort, upgrade risk and total cost of ownership (TCO). 

Building on this, Nova has created an operational management tool for CIOs and CCoE managers with the clean core categorization. Extensions are not assessed across the board, but classified according to the target image: tolerable, to be adapted in the medium term or to be developed from scratch. Clean Core is thus transformed from an abstract objective into a concretely controllable architecture and governance model. The remaining differentiation core is then completely redeveloped with „AI Build“ - on the basis of the specifications created, which can be refined and adapted beforehand in consultation with the specialist departments. 

In contrast to traditional development approaches, business logic can be clarified and readjusted directly at the functional level. Depending on the target architecture, implementation then takes place directly in the SAP system or side-by-side on the SAP Business Technology Platform (BTP). The Nova platform supports code generation, tests, consistency checks and the preparation of the controlled transfer via the SAP transport system to QA and production systems - the customer is always responsible for triggering the transport requests. A key difference to generic coding assistants is that Nova does not simply superimpose a chat interface over the source code. 

Agentic AI platform for SAP

Together, we have built an agentic AI platform that can actually work directly in the SAP context: search code, understand correlations, reconstruct business logic, make controlled changes, trigger tests and integrate into existing governance mechanisms. It is precisely this combination of SAP architecture knowledge, modern AI technology and an operationally resilient platform that makes Nova Intelligence stand out. The operating and security architecture is also crucial for CIOs and CCoE managers. Nova is provided in isolation for each customer, typically in a dedicated cloud environment with clear separation of network, data storage and access rights. 

The connection to existing SAP systems is made via encrypted connections and uses the users defined by the customer with their roles and authorizations. Customer data and customer-specific code are not used to train models. At the same time, the platform is model-agnostic so that new model generations can be integrated without changing governance, operating model or data sovereignty. The result is an AI platform that meets the security, compliance and governance requirements of large SAP landscapes. Another long-term advantage lies in the Nova AI knowledge core. Nova learns with every use: With every interaction, the platform gains a deeper understanding of each company's SAP landscape - its business logic, its architectural patterns and its standards. Knowledge that previously resided in the heads of a few key developers becomes accessible to the entire organization, and the platform continuously improves.

At the same time, Nova is deliberately not designed as a one-off migration tool, but as a permanent platform for the entire lifecycle of SAP custom code: Analysis, documentation, fit-to-standard, clean-core governance, new development, optimization and continuous further development.

Basis for AI-supported processes

For CIOs and CCoE managers, this is precisely the crucial point. Nova not only addresses the one-off transformation, but also the structural manageability of the SAP landscape for years to come. This results in lower lifecycle costs, better upgrade stability, clearer governance and a more robust foundation for future AI-supported business processes.

Exactly twenty years ago, in April 2006, I started the Hana project on behalf of Hasso Plattner to invent the new architecture that would make SAP fit for the future again. On this basis, SAP has modernized its standard code with S/4 Hana. With Nova Intelligence, we now offer the solution to make the custom code on this powerful platform fit for the future - and thus ready for the era of AI-supported business processes. Nova Intelligence is now being used operationally by customers and is generating strategic benefits. This is precisely why we came together in this special team and founded Nova Intelligence.

(Source: Nova Intelligence)