Whereas laypersons usually associate artificial intelligence (AI) with robots in factory halls, a revolution is creeping into master data management. As a tool for highly specialized tasks, AI opens up many use cases that already show signs of their disruptive potential: From obtaining important findings through machine learning to automating manual tasks and making unstructured data usable.
Our understanding of artificial intelligence is constantly changing because the limits of what we believe machines and computers are capable of accomplishing shifts every day. Currently, science is concentrating on having machines imitate special human abilities, such as controlling movements (robotics), perception of the environment (machine perception), or understanding emotions (affective computing). In the context of master data management, on the other hand, machine learning (ML) is one of the most relevant AI capabilities.
Consistent data thanks to machine learning
With regards to ML systems, particular focus is being placed on data validation. These intelligent systems recognize inconsistent data, including what are referred to as “aberrations,” and there is no need for clear rules to be defined in advance. Master data maintenance in itself is another possible use. Without the use of AI, master data maintenance has functioned as follows to date: In master data tools such as SAP MDG, business rules can be defined that help users enter data and guarantee a high level of data quality. The greatest challenge in this process is that every rule to be applied needs to be clearly definable and derived from experienced users.
Every company has employees who can inspect data sheets and intuitively recognize that something isn’t right, even if they cannot justify it or comprehend it using a defined rule. This is the skill that turns normal users into data experts. With the help of AI, systems can be trained to adapt this skill and recognize inconsistent data through machine learning based on training – and they can do that without having to establish defined rules in advance.
The possibilities which an AI system like that offers can be illustrated clearly using a simple material with dimensions in length, width, and height as an example: Although everybody knows that certain combinations of these values are valid in their own product portfolio and others are not, describing the necessary relation among these three values in a defined rule is an extremely complex or even impossible process. AI and machine learning in particular make it possible to check this very relationship based on the product portfolio in question and not on a specific set of rules. Machine learning systems create an algorithm that learns both consistent and inconsistent model combinations from previous data records. Thus, after a certain amount of training time, it is able to validate new master records and issue warnings if an incorrect entry was made.
Even if the models and the AI algorithm are not trained in SAP MDG itself, such as on an R server, validation processes that take place during the creation of new master records can be completely integrated into the system so they can be executed using the SAP MDG validation framework. Users do not therefore need to differentiate whether the data was validated using a classic business rule or an AI model.
Harmonizing ERP environments: AI can help
In ERP harmonization and transformation projects, both process standardization and data harmonization are particular challenges. The ERP systems have often grown over decades. Various people and partners have extended them, meaning the systems are usually quite heterogeneous both with regards to processes and data. That is why MDM teams are of critical importance for projects like these. MDM teams make sure the master data of the systems are merged and sustainably managed from that time on. In this case, machine learning can play an important role as a supplement to the classic ETL (extract, transform, load) applications, for example, for mapping data. Instead of specifying exact mapping logic (data + rule = mapping), ML applications enable optimized mapping based on training data (data + training = mapping). The actual rules fade into the background in that process. The first ML-based standard software solutions for consolidating customer and supplier data already exist. For harmonizing master data, ML approaches also help make values uniform. One example of this is evaluating transactional supplier data for determining optimum payment conditions.
Master data maintenance with chatbot support
User friendliness and system performance play an important role when it comes to user acceptance of master data applications. Especially where user experience is concerned, MDM applications are compared with private smartphone apps – and usually come out on the losing end. Nowadays, users want to be able to maintain master data with a couple of clicks. A large number of master data fields is being critically questioned increasingly often. Personal assistants based on machine learning make completely new experiences possible for users. They usually use speech-to-text and natural language understanding algorithms.
They make it possible to quickly make entries that normally require several clicks using a simple voice command or a chat entry. If users have questions, personal assistants offer help via buttons, selection fields, and selection values or record user feedback for new functions or performance. An additional advantage is that they make it possible to maintain master data using business language. Users do not have to specifically address the identification key (customizing value) in the process.
AI communities develop new application cases
AI-based master data maintenance is enjoying interest across all sectors that is growing continually. Open AI communities, such as the “Global Community for Artificial Intelligence in Master Data Management” that Camelot founded, serve joint research purposes and promote application case development.
We would like to thank Biagio Clemente for his valuable contribution to this article.
