AIGC Innovative Experiment Integration with SAP An...
Technology Blogs by SAP
Learn how to extend and personalize SAP applications. Follow the SAP technology blog for insights into SAP BTP, ABAP, SAP Analytics Cloud, SAP HANA, and more.
The emergence of Artificial Intelligence Generated Content (AIGC) marks a significant shift in the field of artificial intelligence. Generative AI has profoundly changed the way people process information and will further fundamentally change the way enterprises operate.
As the world's leading enterprise management software supplier, SAP has also actively responded and followed this technological trend and proposed a new AI strategy - Business AI . SAP Business AI aims to deeply integrate AI technology into all solutions provided by SAP and make full use of advanced AI technology, especially Generative AI, to provide users with more intelligent and automated applications and promote digital innovation of enterprises. and intelligent decision-making. SAP BTP ( SAP Business Technology Platform) , as the technical base of SAP AI strategy , provides enterprises with powerful capabilities in four aspects: data management and analysis, function expansion, system integration, and AI artificial intelligence services.
This article aims to present a custom project developed by the Business Technology Platform (BTP) team at SAP Greater China. It will demonstrate how to leverage SAP Analytics Cloud to seamlessly integrate with Generative AI technology. The goal is to develop an AI-driven, user-friendly conversational intelligent analysis application that requires no technical expertise.
AI-driven analysis application
Program Background
Currently, data analysis systems offer self-service applications that enable effective visual exploration and analysis of data. However, a certain level of technical expertise is still required, especially in the face of the constantly changing market environment. The business users require faster, more user-friendly, and fully intelligent analysis applications that do not require any technical skills.
Generative AI possesses powerful capabilities in natural language processing, text generation, code generation, and more. However, it also has certain limitations, particularly in terms of the inaccuracies or false outputs sometimes produced by Generative AI, which may lead to erroneous outcomes. It seamlessly integrates the data analytics capabilities of SAP Analytics Cloud with Generative AI, effectively mitigating the hallucinations of Generative AI. It not only supports intelligent analysis based on natural language, but also generates thought processes with business logic by Generative AI, all while ensuring data security.
Figure_1 : Comparison of enterprise analysis platform and Generative AI
Solution Architecture
It‘s an innovative analysis solution that combines SAP Analytics Cloud with Generative AI. It supports interactive analysis through natural language, translating users' natural language commands directly into visual, business logic-driven data analysis charts and reports. This greatly facilitates user data analysis and understanding, enhancing the efficiency of corporate data analysis and the utilization of data.
Figure_2 : System Architecture
It transforms user questions into prompts and sends them to the Generative AI. The Generative AI, based on these prompts and the requirements of the thought chain, generates AI analytical approaches and steps with business logic, and formulates analytical commands in accordance with the syntax requirements of SAP Analysis Cloud's natural language query interface ( JustAsk / Search to Insight). This information is relayed back to SAP Analytics Cloud. Once SAP Analytics Cloud interprets the commands, it can automatically generate analytical charts with business logic, and integrate these charts into an analytical report for user review.
It has achieved a synergy between SAP Analytics Cloud data analysis technology and Generative AI technology, providing users with deep intelligent analysis applications based on natural language. It allows users to quickly switch underlying data models to adapt to various business scenario needs. At the same time, it can also connect with different Generative AIs to meet the needs of different users.
Intelligent Analysis Application
Intelligent Question & Answering
Traditionally users would extract information by finding the correct data and building the appropriate visualizations. This was time consuming and required BI expertise. Whereas now we can provide a new way to interact with your data in order to find usable information to users in order to enable you to make decisions at the right time
Intelligent question-answering can automatically generate visualization charts and analytical summaries with business logic based on users' questions, providing business users with intelligent analysis without any technical barriers.
Figure_3 : Intelligent Question and Answering
Intelligent Q&A allows users to pose questions in natural language in SAP Analytics Cloud (SAC) and send them to the Generative AI platform. The Generative AI firstly determines whether the user's question is related to data analysis. If it is not, it directly returns the corresponding results. Once determined to be data analysis-related, the Generative AI then judges whether the question is simple or complex to ascertain if the analysis requires one or multiple steps. The Generative AI, following the settings required by the business thinking chain, provides an AI analysis approach with business logic and continuous related analysis steps, and generates the corresponding SAP Analytics Cloud commands. SAP Analytics Cloud will then automatically generate a series of visual analytic charts with business logic based on these commands, and generate corresponding analysis summaries following the analysis approach. The number of steps in the thinking chain, business logic, the quantity, and types of charts, etc., are autonomously determined by the Generative AI.
It's worth mentioning that, considering users' concerns about data security, we have specifically set up related data security protection mechanisms. In the Intelligent Q&A, users can choose not to send their data to the Generative AI to ensure data security. Even in the mode of not transmitting data, it can still generate AI analysis approach and analysis charts based on metadata.
Intelligent Reporting
Intelligent Reporting is based on Intelligent Q&A and adds the function of automatically setting the analysis report page. Intelligent Reporting can automatically determine the analysis ideas and steps according to the user's open analysis needs, such as "make an income analysis report", and combine it with the business thought of chain to automatically generate a one-page analysis report. The number of charts, types of charts, and page layout in the report will be automatically set according to the AI analytical approach of the Generative AI.
We know that Generative AI can understand natural language and respond in a natural language manner. They can also understand and generate very complex texts. In the project, we fully leverage the natural language processing capabilities of Generative AI to understand users' data analysis needs. Users can pose their questions in a completely natural language, including complex ones. The Generative AI can fully comprehend these questions and translate them into the necessary analysis instructions for visual data analysis.
Simultaneously, we harness the text-generation capabilities of the Generative AI, allowing it to auto-generate analysis summaries that follow human language habits and logic based on chart information and analysis requirements. The summary content can include completion status of key performance indicators, operational issues, operational analysis conclusions. The generated content is returned to SAP Analytics Cloud to provide the users with a combined graphic-text analysis report.
It enables users to conduct data analysis in a natural language interactive manner, making complex data analysis tasks more concise and efficient. This greatly enhances the efficiency of data analysis and realizes zero-threshold intelligent analysis applications.
Few-shot Prompt improves accuracy of Generative AI
Few-shot Prompt is a technique that allows Generative AI to learn and predict with a small number of training samples. By providing a small number of input and output examples in prompt, we can guide Generative AI to produce desired outputs based on new inputs.
In the project, we’ve set a small number of prompt words according to the grammar standards of SAP Analytics Cloud Natural Language Query function. This approach allows the Generative AI to learn from grammar examples and generate accurate analytical instructions in accordance with grammar standards, even without a large amount of annotated data. With SAP Analytics Cloud Natural Language Query function, users get quick answers to questions and incorporate these insights into a story while working with indexed models based on acquired and SAP HANA, SAP S/4HANA, SAP Universe, and SAP BW live data.
As a result, it can effectively reduce the hallucinations of the Generative AI and prevent the Generative AI from producing misleading outputs.
AI thought of chain setting
It is not only to generate an analysis chart, but to generate a series of analysis charts based on business logic to provide complete answers to analysis questions.
In the project, we’ve set it up according to the 'chain of thought' method used by the Generative AI, allowing it to transform the user's problems into a series of connected business thoughts, and the business analysis charts and reports. The number of steps in the thinking chain is determined by the Generative AI based on the understanding of the problem.
Figure_6 : Business thought of chain
(The customer's question is "Identify the two industries with the lowest income and conduct an in-depth analysis". It proposed five suggestions with logical continuity. )
The business thinking chain makes full use of the powerful understanding ability of Generative AI for business analysis. To a certain extent, it can supplement the existing business analysis capabilities of analysts. Taking it a step further, if it is paired with fine-tuned AIGC models that possess industry knowledge and combined with real-time enterprise data, users can obtain highly professional intelligent business analysis insight.
Enterprise Data Security
The Generative AI itself does not have a mechanism for data access management. Based on SAP Analytics Cloud and the underlying SAP data platform, we can set user permissions in the data model. It can implement fine-grained data access control: for the same question, it can generate different text and charts based on the set user permissions, achieving content generation access control. This ensures that each user can obtain information within their permission range, safeguarding data security while also improving work efficiency.
Figure_7 : it generates different content for different users to the same question.
Multi-language support
It can seamlessly connect data models in various languages, and can also understand and process data in the language of the questioner, transform the data into clear and easy-to-understand answers in the user's language , and output analysis results, providing great convenience as an excellent platform for information exchange.
Figure_8 : English and Japanese analysis results
open to various Generative AI platforms
It allows users to openly and conveniently access different AIGC models, meeting a wide range of user needs. At the same time, it also supports users in quickly switching underlying data models to accommodate various business scenario requirements.
Summarize
The project effectively harnesses the power of Generative AI technology and combines it with the robust data analytics capabilities of the SAP BTP platform to enable interactive data analysis through natural language. It also utilizes AI business thought chains to generate analytical reports with business logic. While ensuring data security, it helps users to achieve intelligent insights with zero technical expertise, significantly improving data analysis efficiency and data utilization, quickly responding to business needs, and providing strong decision-making support for enterprise development.
Co-author:
Chang Junling BTP Senior Solution Architect, SAP GC
Yin Haining BTP Centre of Excellence, SAP GC
Disclaimer: SAPnotes that posts about potential uses ofgenerative AI and large language modelsare merely the individual poster’s ideas and opinions, and do not represent SAP’s official position or future development roadmap. SAP has no legal obligation or other commitment to pursue any course of business, or develop or release any functionality, mentioned in any post or related content on this website