Potential for analyzing and increasing the efficiency of AI tools
Every area of private and public life is currently being examined for its suitability for AI. It is hardly surprising that “artificial intelligence” is being touted as a comprehensive savior. Fueled by the impressive possibilities of large language models (LLM) such as ChatGPT, depending on who you talk to, you encounter a mixture of vague fear and absurd expectations. Werner Güntner, managing director of SoftconCIS GmbH and an expert in purchasing control, has examined the topic from the perspective of purchasing control and provides answers.
Procurement today: What expectations are appropriate for supporting strategic purchasing control?
Werner Güntner: Much of what is currently referred to as AI is actually based on rule-based programming, in which data is processed by software rules to achieve results. Artificial intelligence, on the other hand, is used when no clear rules are known. Despite the large amount of data available, it often remains unclear how the data input is related to the result. The idea behind AI models is to simulate known results from data input until the trained model is able to make decisions independently. For example, a language model such as an LLM generates its responses based on the mass data fed into it. What initially reads incredibly well is ultimately based on statistical correlations linked to certain keywords.
Does AI offer new potential for improving purchasing control?
In cases where topic-related mass data is available that enables a model to be trained, the use of AI for purchasing controlling and purchasing management can save a considerable amount of time.
What purchasing topics are conceivable?
The use of an AI model is particularly suitable for the automatic review and improvement of product group assignments, as manual categorization is usually insufficient. However, it is first necessary to train the AI model using a training model with product group assignments that are as accurate as possible, based on suitable material master data. The more precise and therefore meaningful this learning model is, the more successful the AI will be in checking and making future automatic assignment recommendations. Our investigations in the WebCIS modules have shown that automatic assignments and correction recommendations with a quality score of over 75% are possible.
How can AI-based price benchmarks and renegotiation recommendations help identify renegotiation potential for product groups and suppliers?
Improving quality allows more homogeneous clusters to be formed, resulting in more accurate price benchmarks both within product groups and in comparison to external price indices. The classic question asked by purchasing managers, “In which product groups and with which suppliers is there potential for renegotiation?”, can now be answered automatically.
How do similarity analyses support purchasing?
Similarity analyses help to identify technically comparable parts. The aim is to reduce unnecessary variety and highlight unjustified price differences. AI is used to create similarity scores based on the technical characteristics of the material descriptions. These values show how similar materials are and provide helpful and meaningful ideas for purchasing decisions.
How can AI help with supplier price demands?
The question of the extent to which prices and price change demands from suppliers are in line with the market is as old as purchasing itself. Cost breakdown is considered a common approach to price evaluation. It is also known as cost structure analysis and is a detailed breakdown of the cost components of a product or service. But couldn’t AI-supported pattern analysis also be helpful? Price components are features of a product that, when weighted differently, make up the total price.
But how exactly does each feature contribute to the price? AI models simulate the characteristics that influence the price until mathematical patterns become apparent. These multivariate regressions are based on statistical correlations rather than technical effects and offer the advantage of providing insights automatically and quickly. However, like all AI applications, they depend on the quality of the underlying learning model.
What other applications do you see in the supply chain?
What has long been part of everyday life in banks and insurance companies – namely, using AI models to classify the creditworthiness and risk assessment of borrowers – should also be considered as an additional approach in purchasing. Purchasing has sufficient features – supplier evaluation criteria, external scores, audit results, country scores, product group benchmarks, price indices, sourcing information, etc. – to feed a learning model that highlights warnings about the supply chain. In addition to an analytical risk assessment with weighted influencing factors, this allows potential problems in the supply chain to be identified at an early stage.
What do you think about using WebCIS/LLM or chatbot prompting for optimal argumentation suggestions?
A comprehensive information system such as WebCIS AI offers versatile functions for evaluations, reports, and dashboards that are specifically tailored to the needs of purchasing. With structured dashboards, customizable favorites, and reports that can be created on the fly, almost any question in the area of purchasing can be answered. The key question, however, is which feature is best suited for a specific issue. The integration of a large language model (LLM) into the WebCIS interface makes it possible to obtain suitable suggestions through targeted prompting. An advanced form of analysis can also pass filters on to the corresponding functions. The advantage for the user lies in the greater variety and depth of information, combined with simplified operation, as continuous text is intelligently processed.
Published in Beschaffung aktuell » 05 | 2024 Page 32/33
