Accelerating SAP consulting projects with SAP Joule using Anthropic Claude models on Amazon Bedrock

This post was jointly authored by AWS and SAP. We thank Sachin Kaura from SAP Joule For Consultant team for their contributions to our successful partnership and this blog post.

Introduction: Challenges faced by consultants

As SAP consultants increasingly work on complex cloud transformation projects, accessing critical implementation guidance and best practices poses a challenge. Consultants often spend valuable billable hours searching through extensive documentation, certification materials, and SAP Knowledge Base articles to find the right information for specific client scenarios. Key implementation steps, configuration details, and troubleshooting guidance are found across multiple sources, which means consultants spend time finding technical documentation when they should be focused on delivering value to clients. This causes delays in project delivery and increased risk of rework, as teams struggle to meet deadlines during critical implementation phases.

To address these challenges, the SAP team partnered with AWS to develop SAP Joule for Consultants (J4C). Using Anthropic’s Claude models via Amazon Bedrock, J4C provides natural language access to SAP’s most authoritative and comprehensive knowledge. This innovation allows consultants to quickly navigate and understand technical implementation requirements through conversational AI interactions with authoritative SAP content. J4C helps consultants translate the business requirements into concrete implementation.

What is SAP Joule for Consultants?

SAP Joule for Consultants is the conversational AI feature of the SAP Joule assistant that helps consultants accelerate SAP projects with expert guidance using SAP’s most exclusive and up-to-date knowledge base. This consulting capability of the Joule assistant is designed to boost consultant productivity by providing fast, reliable answers grounded in exclusive SAP content, helping with design decisions and ensuring alignment with best practices.

How Joule for Consultants empowers SAP consultants and partners

With SAP Joule for Consultants, consultants can quickly assess solutions for specific business scenarios and access implementation best practices. The feature transforms lengthy documentation into conversational, accessible guidance without requiring extensive searching, thereby significantly reducing time spent during project delivery. Consultants, including both junior and expert practitioners, benefit from improved efficiency while working on client engagements, while expert knowledge becomes more accessible in real-time contexts. As a result, they spend less time searching for information and documentation and more with customers and personal interactions.

Laying the foundation: from vision to first deployment

After Amazon Bedrock’s integration with SAP’s Generative AI Hub was announced featuring Claude models, AWS initiated collaborative workshops with SAP’s Joule team, identifying SAP’s vast repository of certification materials, knowledge base articles, and community content as prime candidates for Generative AI enhancement.

The project began with creating a comprehensive evaluation dataset to ensure quality and accuracy. The team leveraged SAP’s existing material via the extensive certification catalog from learning.sap.com/certifications. This approach made use of already existing structured question-and-answer pairs, which allowed creation a comprehensive evaluation dataset. The key breakthrough was developing a completely automated data driven evaluation pipeline using the question and answer pairs. With this approach, the team could quickly and systematically benchmark several candidate models and retrieval-augmented generation (RAG) implementations.

The team experimented with a series of prompt engineering experiments to optimize model performance. For example, they discovered that using authoritative language like “You are an expert SAP test taker” produced significantly better results than tentative phrasing such as “You are acting as a SAP test taker.” These refinements, guided by continuous testing against the certification dataset, helped establish best practices for consultant-focused interactions. Furthermore, the team discovered that real world Q&A was more effective than using intentionally tricky test questions for the evaluation dataset.

Architecture Deep Dive: Retrieval Augmented Generation as a backbone

The “Joule For Consultant” project focuses on utilizing retrieval-augmented generation (RAG) pattern to provide consultants with precise answers to their queries directly from SAP certifications learning material, SAP KBAs, SAP Notes. Certification material was multi-modal – meaning it contained text, images as well as charts which Anthropic Claude models excel in understanding.

The RAG system included a data ingestion and processing pipeline tailored to SAP’s extensive certification knowledge repository. Structured and unstructured documents in SAP certification materials were systematically collected, cleansed, and chunked into a retrievable format. Key steps include extracting text, metadata (e.g., SAP product versions, applicability, implementation phases), and resolving cross-references between documents. Preprocessing techniques, such as semantic chunking were applied to segment content into contextually meaningful units while preserving SAP-specific technical nuances. These chunks were then encoded into high-dimensional vector embeddings for the RAG implementation. The processed data was indexed in the HANA vector engine optimized for fast similarity searches, ensuring the system can efficiently map consultant queries to the most relevant SAP content during retrieval.

During the query phase, the RAG service combines retrieval and generation to deliver precise consulting guidance. When a consultant submits a query, the system first leverages the vector database to retrieve the top SAP knowledge snippets semantically aligned with the query intent. A reranking step prioritizes results based on relevance scores, content recency, or applicability criteria to ensure up-to-date and actionable insights. The retrieved context is then fed into Amazon Bedrock’s Claude models via SAP’s Generative AI Hub, which synthesizes a concise, natural language response tailored for consulting scenarios. Crucially, the response is grounded strictly in the retrieved sources, with built-in citations to original SAP documentation for transparency and further exploration. This hybrid approach balances speed and accuracy, allowing consultants to resolve implementation challenges in real time while minimizing hallucinations, thereby reducing reliance on traditional support channels and accelerating project delivery.

Customer Value and Benefit

According to Sachin Kaura from SAP Business AI and Chief Architect of SAP Joule for Consultants:

“SAP Joule for Consultants is built with SAP’s authoritative knowledge base—spanning more than 25 million documents and 12 terabytes of SAP expertise data — delivering clear, cited answers in seconds. By leveraging Anthropic Claude models on Amazon Bedrock, we’re able to scale this capability for SAP’s global consulting ecosystem, supporting consultants at every level and helping teams deliver higher-quality SAP projects up to 14% faster for customers.”

For additional customer testimonials and insights, visit the SAP Joule for Consultants page.

SAP’s initiative to enhance the consulting experience through SAP J4C provides significant business value across the entire SAP partner ecosystem. The business impact extends beyond individual consultant productivity. With SAP’s partner ecosystem, any improvement in consultant efficiency translates to massive economic benefits. SAP J4C saves consultants up to 1.5 hours per day through instant access to trusted SAP knowledge, significantly compressing SAP project timelines.

Conclusion

SAP Joule for Consultants demonstrates the transformative impact of thoughtfully applying generative AI to enterprise consulting challenges. Leveraging the integration of Anthropic’s Claude models on Amazon Bedrock through SAP’s Generative AI Hub, the team addressed the fundamental problem of knowledge accessibility by implementing conversational AI capabilities for SAP’s vast repository of consulting knowledge. This solution addresses the critical need for efficient access to authoritative SAP guidance during complex implementation projects.

We encourage you to explore SAP Joule for Consultants and see firsthand how Generative AI can transform your consulting practice and project delivery capabilities and review Amazon Bedrock and Anthropic Claude pages to dive deeper into these technologies.