LLMs massively expand the possibilities for AI-supported language processing. Large language models use deep learning techniques and are trained to recognize the patterns and relationships between words and phrases using billions of data sets. This enables them to process, understand, and generate human language.
Another group of digital purchasing assistants are LLM negotiation bots for tail spend orders. Due to the low individual volume of tail spend orders, most purchasing organizations do not consider it economically viable to use their own employees for negotiations. However, although individual tail spend orders have a low cost, the total volume can accumulate considerably.
Another group of digital purchasing assistants are LLM negotiation bots for tail-spend orders. Due to the low individual volume of tail spend orders, most purchasing organizations do not consider it economically viable to use their own employees for negotiations. However, although tail-spend orders individually cause low costs, the total volume can accumulate considerably.
LLM negotiation bots can exploit the savings potential of tail spend orders by automating the entire negotiation process: from selecting the orders to be negotiated, determining the negotiation strategy, conducting negotiations with the supplier, up to implementing the order.
Determination of product groups
A third current example is innovative LLM bots for automatically determining the product group, as provided by SAP Buying 360—a new function of SAP Ariba Buying and S/4 Cloud, Public Edition. In this use case, the use of AI also greatly relieves the burden on retailers, as determining merchandise categories is one of the most time-consuming tasks in their day-to-day work.
LLM bots offer purchasing departments a number of key benefits. In addition to tail spend savings, the reduction in time and costs through automated operational procurement processes ranks right at the top. As monotonous, repetitive tasks are taken over by machines, retailers can concentrate on strategic and innovative tasks. This increases their motivation and allows the sales department to focus more on improving procurement processes and supplier management. Regulatory requirements such as the German Supply Chain Duty of Care Act (LkSG) and increasing supply bottlenecks require comprehensive supply chain monitoring and risk management, now more than ever.
Whereas requesters would have to rely on manual support from often overworked retail staff, the FAQ and negotiation bots can access the backend systems directly and can deal with any query or request immediately. In addition, the bots also optimize purchasing data quality. In order for the LLM bots to work reliably, the data records in the SAP back-end systems must be up-to-date, complete, error-free, consistent, and unmistakable. This forces retailers to continuously maintain their data. Requesters must also pay attention to these quality criteria when entering data to ensure their requests can be answered satisfactorily. High data quality significantly reduces error rates in operational procurement processes.
The introduction of modern LLM bots in SAP-based purchasing organizations should follow a clear road map. The next step is to use prompt engineering methods to communicate these questions to the generative AI model in such a way that it delivers high-quality and relevant results. As large language models can be trained to immediately recognize the intention of the questioner and assign the question to the corresponding use case, the development and implementation effort is significantly lower than for conventional chatbots. This is because separate paths have to be built for each use case. In contrast, a single LLM bot can cover many different tasks and applications within a retail organization.
It is important to start with change management measures early on in the project in order to increase user acceptance. This is the only way to ensure that requesters and suppliers are prepared to interact with the new text robots once they are released and put into operation.