Artificial intelligence is being touted – appropriately or not – as a future panacea for every conceivable economic challenge, supposedly offering out-of-the-box solutions to every conceivable question.
This raises the question of what “artificial intelligence” actually means, how it differs from analytical calculations or digital assistance, what WebCIS 4.0 already includes today, and what can be expected in the future.
Analytics and algorithms
In any software based on fixed program code, data input is processed predictably according to fixed rules using algorithms to produce output.
In purchasing controlling, calculations are usually based on analytical software solutions, which are primarily intended to support reporting.
WebCIS 4.0 offers the following examples: Determination of material cost development, currency effects, maverick buying quota, P&L impact through the use of payment terms, plan/actual (forecast) comparisons, calculations of supplier quality and delivery reliability, etc.
Digital assistance for purchasing control
Since many of the decisions to be made in purchasing are based on human interactions, the lack of high-quality learning characteristics means that it is rarely possible to propose an optimal solution, which is a key goal of artificial intelligence.
The digital assistant integrated into WebCIS 4.0 offers a solution by means of correlations that relate two or more characteristics to each other. Correlations and regressions show connections in the form of bubbles or point clouds that cannot be identified analytically. Trained by purchasing expertise, the digital assistant provides strategic purchasers with ideas for proactively examining the product group environment for potential actions.
Possible applications of the digital assistant in WebCIS 4.0 to support strategic purchasing include: Do price and quantity developments match, are correlations between delivery reliability and quality recognizable as early indicators, do order and invoice prices deviate from each other and are there significant dependencies, etc.?
“Artificial intelligence” or machine learning?
“Artificial intelligence” (AI) often evokes false associations. AI is merely an umbrella term that encompasses topics such as supervised learning, unsupervised learning, and neural networks.
AI is therefore a far cry from what we understand as human intelligence. Since it is more a matter of software systems and patterns learned from them, the term machine learning (ML) is more appropriate.
Machine learning simply means using mass data to train the output of a software program to focus on a very specific topic by providing it with the highest quality data input possible.
In purchasing, the question often arises: Is a supplier a “preferred supplier”? This issue can be addressed using a questionnaire or by training ML-based software to identify which existing suppliers are “preferred suppliers” based on their data characteristics. The ML module does not use algorithms to calculate what a “preferred supplier” is, but rather uses defined input data to generate independent rules that indicate which feature correlations could indicate a “preferred supplier.” The quality of the recommendations made by the software is expressed in terms of probabilities. There is no longer right or wrong, there is a more or less suitable recommendation. This is a circumstance that ultimately makes humans the decision-makers, not the software alone.
One area that is very well suited for machine learning, when docked to WebCIS 4.0, is the assignment of product groups to material numbers. Well-trained software is highly capable of suggesting an assignment for a new material or proposing the restructuring of an existing assignment. Assigned a probability, humans can assess how well-founded the machine’s suggestion is.
Another area of application for ML is in the context of maverick buying, which is transparently displayed in WebCIS 4.0, where material numbers and thus product groups are usually missing. The expectation of the ML-trained software docked to WebCIS 4.0 is to automatically deliver a suitable product group assignment based on trained booking texts, G/L accounts, and suppliers.
