With ChatGPT at the latest, generative AI (artificial intelligence) is on everyone’s lips and is finding its way into more and more processes and workflows in various areas and industries.The term “generative AI” refers to a technology that can generate new content based on existing information and specifications from a user. Its associated potential is becoming increasingly apparent with the availability of trained generative AI models and Large Language Models (LLM) to the general public.
Increased productivity and innovation, higher effectiveness, optimized quality of results, better decisions, and reduced costs are just a few of the benefits of using AI. However, how to incorporate generative AI into a company's operations, including their (SAP) systems and software environments? In this blog, we shed light on possible use cases.
A technology that in the past was only accessible to researchers, IT developers, and mathematicians is now usable via a single search line. The results make people speechless and excited. ChatGPT has created a “Google moment” that hints at the full extent of potential and possible use cases that come with generative AI.
Generative AI is based on machine learning models modeled generatively and trained using mass data to produce new data as similar as possible to the training data set but not identical. Thus, these models are probabilistic rather than deterministic. The goal is to generate as many variations of a training data set as possible that have a high probability of matching that data set. Deterministic models, on the other hand, always generate the same results for specific frame parameters based on manually assigned descriptions, labels, or tags. Labeling and tagging cause high manual effort for mass data, so deep learning approaches such as neural networks or deep neural networks (DNN) are used to autonomously partition the training data into different dimensions and thus label or tag them.
There is often discussion about whether generative AI will replace human work. Consequently, many employees perceive the technology as a threat, ignoring the added value of artificial intelligence: Automating repetitive tasks by integrating AI into daily business processes, leaving more time for creative, communicative, and social tasks. Not only does it result in an immediate increase in effectiveness and efficiency, it also increases innovation. Likewise, it results in cost savings and quality increases since AI support reduces the workload on the one hand and, on the other hand, prevents errors or improves error detection.
Generative AI and Large Language Models play a crucial role here, as they do not provide rigid results to a specific input like previous deterministic machine learning models but various results and possibilities. When training generative AI and LLM with a large amount of data with different contexts and multiple dimensions, such models can learn complex relationships and dependencies. For example, they can render languages grammatically correctly or suggest solutions to various questions related to the topics included in the training set. OpenAI has impressively demonstrated these capabilities with ChatGPT and Google with Bard.
How can such trained and mature models be adopted for business processes in a company and integrated into (SAP) software and systems? We see four options for integrating and using generative and pre-trained AI models (we exclude fully custom-trained models):
Available models can be deployed immediately with their current capabilities.
Available models get access to specific in-house data to apply their capabilities to this information.
Available models are trained with rules and dependencies (small data sets with high data quality) to use their existing capabilities within the defined scope and set of rules.
Available models are trained and adjusted explicitly for specific use cases (large data sets with moderate data quality) to provide accurate results for the selected use case.
The options listed are not mutually exclusive, but can be related to each other.
As far as the possible use cases of generative AI in the enterprise are concerned, seven overarching, domain-independent, and system-unspecific scenarios can be distinguished:
A trained model such as ChatGPT can serve as a “Natural Language Interface” within software and systems such as SAP S/4HANA, allowing users to interact with the system via simple conversations or chat. Users can ask questions, request information, or perform transactions and operations conversationally, making interaction with the system simple and intuitive. Similarly, input can be validated against predefined criteria, rules, or quality specifications, ensuring data accuracy and thus reducing error rates.
Following the concept of an intuitive language interface, generative AI models can be used as virtual assistants or chatbots to provide support and answer user queries within software and systems. For example, they can help users find information and troubleshoot problems or actively guide them through processes, reducing reliance on traditional support channels and improving user experience.
This scenario helps users actively analyze data in software and information systems. Users can ask questions about specific data points, request summaries and visualizations for data, or gain insights from system and software data. Thus, users can quickly access relevant information and make data-driven decisions. For example, by providing real-time access to key metrics or report summaries, users can ask questions about specific data and business areas, request custom reports, or receive automated updates on business performance.
In this scenario, generative AI provides interactive tutorials and customized support to users new to a system or software. Integrated or as an external chat bot, it can issue step-by-step instructions, answer user questions, or provide contextual information to help users navigate processes and find their way around the system more quickly.
Another scenario is to build an intuitively searchable knowledge base or documentation repositories within software and systems such as SAP S/4HANA. Such a solution provides information about system functionalities, process guidelines, or best practice approaches on demand or proactively when using the system. Users can also access relevant documentation via simple conversations, search for specific topics, or request explanations of system functions. Similarly, it generates interactive and personalized training modules, simulations, or quiz questions to help users learn about system features.
This scenario forms the basis for the previously mentioned “Interactive training and onboarding”. It also serves as a valuable knowledge transfer tool for sharing the know-how of experienced users or experts. Likewise, integrating existing knowledge databases such as Atlassian Confluence or Microsoft SharePoint makes it possible to improve search functions, find relevant and associated documents more easily and quickly, and identify user intentions and perform context searches on them. This integration can save search time and increase user productivity.
Generative AI can leverage user preferences, historical (usage and query) data, and system knowledge to provide personalized recommendations to proactively offer relevant reports, drive process improvements, or deliver insights based on user-specific contexts. Considering such information helps users optimize their workflows and decision-making.
By incorporating predefined trigger points or user preferences, a generative AI solution can notify users proactively of critical events, system updates, or anomalies detected in data before, during, or after they occur to enable rapid action and early countermeasures.
With the increasing availability of proprietary trained generative AI and LLM models such as ChatGPT from OpenAI or Bard from Google, the barrier to entry for integrating generative AI into business processes in one’s own company has fallen massively, and the number of possible use cases has risen sharply. Specialized AI researchers and developers are no longer required to create custom topic-specific machine learning models. However, using, training, and adjusting generative AI and LLM models still requires expert knowledge.
Generative AI models can be used and integrated out of the box, significantly reducing AI projects’ complexity, effort, and cost. The know-how requirements for users are also significantly lower (to a certain extent), as queries can be made in their language style and according to their understanding, and no fixed query syntax is required, which reduces the development and training effort required to cover widely varying queries. As a result, the complexity and scope of AI integration projects have decreased, making it possible to introduce AI-supported functionalities quickly and easily compared to previous projects. SAP has recognized this and increasingly offers AI-supported functional modules for solutions such as SAP S/4HANA or SAP Analytics Cloud. In our next blog, we shed light on SAP’s strategy for generative AI and which SAP solutions already include AI components.