Artificial intelligence in controlling - an outlook

KI wurde das erste Mal 1956 auf einer Konferenz am Dartmouth College erwähnt. Während sich die Erwartungshaltung in den darauffolgenden Jahren in Grenzen hielt und viele Forschungsprojekte im Rahmen sogenannter KI-Winter sogar ganz eingestellt wurden, hat das Thema in der jüngeren Vergangenheit enorm an Dynamik gewonnen und ist zum Megatrend geworden. Viele Anwendungen spielen bereits seit einiger Zeit eine wichtige Rolle im Leben vieler Menschen. So basieren beispielsweise Suchmaschinen, Social Media Feeds und Spamfilter in E-Mail-Diensten auf KI. Auch die aktive Nutzung ist relativ einfach geworden und eine dezidierte Ausbildung ist in vielen Fällen nicht nötig, um brauchbare Ergebnisse erzielen zu können. Stichwort: ChatGPT. Bereits heute kann KI Geschäftsprozesse verbessern und beschleunigen. Zahlreiche Unternehmen setzen entsprechend auf die Technologie. Der Anteil dieser Firmen liegt laut einer Studie des ifo Instituts in Deutschland derzeit bei 13,3 Prozent. 

While the underlying processes for the way AI works are not new, there are some developments that have paved the way for progress in recent years. On the one hand, companies now have access to large amounts of data (big data). Secondly, there are optimized AI algorithms that can be used to make better use of this data. In addition, the computing power and capacity of computers has increased significantly in recent years. In line with Moore's Law, the number of transistors installed on a chip has doubled approximately every two years since 1965. 

What is AI?

Doch was ist KI eigentlich? Der Begriff beschreibt selbstlernende Computersysteme, welche durch Nutzung von großen Datenmengen und Algorithmen menschenähnliche Intelligenzleistungen wie Problemlösen oder Lernen erbringen, den Menschen bei der Entscheidungsfindung unterstützen und ihn schlussendlich auch ersetzen können. Dabei gibt es unterschiedliche Entwicklungsgrade, von Algorithmen, die repetitive Aufgaben übernehmen, bis hin zu Big-Data-Analysen und selbstlernenden Systemen. 

The purely mathematical processing of data sets with basic arithmetic operations is also considered AI in some definitions. However, such methods are more likely to be found in the field of data analysis and statistics than in AI and fall less within the specific scope of AI functionalities. 

The way an AI works can be described in the following steps. 1. the acquisition of structured and unstructured information and data similar to the process of human sensory perception, 2. the analysis and meaningful processing of the acquired information and data and 3. the implementation of actions based on the acquired information. A further, fourth step is independent learning through training and feedback based on the previously collected data. The last step in particular is important for the existence of an AI and distinguishes it from other applications, some of which are already complex. 

Areas of application for AI

The main areas of application of AI in companies that are already relevant today can be divided into different categories. 

On the one hand, there are human-to-machine dialog processes. These offer the opportunity to communicate with the machine or computer in natural language in order to avoid complex screen or keyboard interactions. This can take place in verbal or written form. One example of such dialog processes is the voice control of navigation systems or virtual voice assistants such as Amazon Alexa. 

On the other hand, there are machine-to-machine processes, which are based on the networking of technical devices with each other and with a central logic. The data exchange made possible by this is essential for applications in the Internet of Things (IoT), for example. One specific area of application is the prediction of upcoming maintenance work using machine learning processes that draw on sensor data. This can be done, for example, with air conditioning systems as part of building control. 

Intelligent automation (IA) describes the combination of process optimization with AI. It helps companies to improve their internal processes. One of the many areas of application is the automotive industry. Here, IA can be used to predict and adapt production more effectively in order to respond better to supply and demand. Furthermore, robots and automated systems can take over tasks such as assembly, welding, painting and quality control, thus minimizing employee injuries and providing higher quality products at lower costs. 

Die letzte Kategorie, die intelligente Entscheidungsunterstützung, beschreibt die Analyse von Daten mit KI-Algorithmen, um effektiver Entscheidungen treffen zu können. Anwendungsfelder sind beispielsweise Assistenzsysteme in der Medizin, wo KI-basierte Diagnostik die Menschen unterstützen kann. Damit KI bei der Entscheidungsfindung der menschlichen Anwender*innen wirkungsvoll unterstützen kann, bedarf es Daten in hoher Qualität. 

How well developed is AI in controlling today?

Der Einsatz von KI im Controlling kann unterschiedliche Ausprägungsformen haben, wobei nicht alle davon schon heute eine gelebte Realität in Unternehmen sind. Die Ausprägungsformen lassen sich in die vier untenstehenden Stufen einteilen. 

Die erste Stufe ist die semi-intelligente Datenanalyse, welche in vielen Unternehmen genutzt wird, um die menschliche Intelligenz zu ergänzen. Konkrete KI-Techniken, die in dieser Stufe des KI-Einsatzes genutzt werden, sind automatisierte Datenbereinigung und -vorbereitung, Mustererkennung und Prognosen sowie automatisierte Finanzberichterstattung und -analyse. 

The next higher level of development involves proactive support from AI based on an even more comprehensive data basis. A corresponding intelligent assistance function can support controllers in performing their tasks in various work situations. This can be done, for example, in the context of interpreting complex financial data, generating real-time reports or assisting with budget preparation. 

The content-enhanced use of AI represents the third stage of the forms of AI use in controlling. Here, AI can not only interpret data, but also make context-related recommendations for controlling measures based on company-specific goals and specifications. In this form, the AI has extended decision-making autonomy and can perform certain repetitive tasks independently or make smaller decisions based on recurring patterns in the data. 

Die vierte Stufe umfasst den umfassenden strategischen KI-Einsatz. Die KI ist hier in der Lage, Daten autonom zu analysieren und auf Basis der Ergebnisse strategisch zu handeln oder eigenständig strategische Maßnahmen vorzuschlagen. Sämtliche Ansätze der vorigen Stufen werden hier kombiniert. Die wichtigste Lernquelle für die KI in dieser Stufe sind Verhaltensmuster von Controllern, welche durch das KI-System beobachtet werden. Es kann so alle relevanten Ursache-Wirkungs-Zusammenhänge im Controlling-Kontext des jeweiligen Unternehmens beobachten und aus ihnen lernen. 

Use cases from the first two stages can already be found in companies today. However, AI has not yet reached the stage where it can independently identify optimization options and name the most efficient approach. This means that the technology is still a long way from the fourth stage mentioned above. 

Examples of AI in controlling that are already relevant today

The first two levels mentioned above offer various fields of application for AI in controlling with different levels of benefit for stakeholders. 

One of these areas of application is forecasting. By evaluating past data, for example on supply, demand, sales figures or production costs, AI tools can integrate various historical data streams. If this data is modeled forward in time, reliable forecasts can be created for numerous areas of the company. 

In addition to forecasting, the planning process can also be made significantly more efficient with AI systems. They can be used to find patterns and abnormalities in transactions, financial data and annual reports and identify drivers, for example for sales planning. In the planning process in particular, it makes sense to also use external data such as economic indicators in order to better assess developments in the market, competition or risks, for example. 

AI can also help with risk management in controlling. By processing data from various sources and recognizing patterns, AI algorithms can identify risks at an early stage. For example, modern tools are able to review contracts and ensure compliance with applicable regulations. 

AI can also help with data analysis . Thanks to AI, this can be largely automated without the need for large human resources to provide information. Data from various functional areas such as sales, production or logistics can be quickly evaluated and immediately presented in graphical form with the help of AI. 

What should be considered when using AI in controlling

For the embedding of AI in a controlling department to succeed, certain basic requirements must be met in the company. If the company has problems with generating, processing and completing data or is struggling to set up a seamless digital tool landscape, the introduction of AI will be difficult. An implementation should therefore only be considered if a comprehensive database and relevant tools are available and the processes in the company are prepared for the change. The use of an AI-based system in controlling only makes sense if, for example, standardized processes and a complete database have been created by using an ERP system. 

It should also be noted that when working with AI, there is a human component that formulates data models and selects algorithms. This is because, at least at the current stage of development, input from a human is necessary so that the data queries result logically from everyday business and correlations are not incorrectly interpreted as causalities. At the same time, humans can ensure that the data is comparable and consistent. The symbiotic interaction between AI and humans therefore currently offers the greatest opportunities to increase effectiveness in the company and achieve competitive advantages. 

A clear objective should also be defined before AI is used in controlling. The technology can only be successfully integrated into the various stages of the value chain if it has been precisely formulated what the use of AI in controlling is intended to achieve. Possible goals could include reducing the probability of errors, improving the quality of decisions or providing information more quickly. 

In order to achieve such goals, a roadmap should also be drawn up. Questions that controlling managers should ask themselves are How can I prepare my data well for working with AI? How can I extract information from the data? How can I derive conclusions from the data? And how can I use these conclusions and external data to create models? 

Conclusion

Der Einsatz von KI verändert die Prozesse in Unternehmen nachhaltig. Das gilt auch für das Controlling. KI bietet zahlreiche mögliche Anwendungsmöglichkeiten für den Bereich, die teils schon heute genutzt werden. Sie erleichtert nicht nur die tägliche Arbeit, sondern liefert auch neue Einblicke. Dafür ist in vielen Fällen weder ein Data Scientist, noch ein*e KI-Spezialist*in nötig. Die Frage nach der Nutzung von KI ist für die meisten Controllingabteilungen konsequenterweise nicht eine des Ob, sondern des Wann. Führungskräfte sollten dementsprechend frühzeitig beginnen, sich mit der Technologie zu beschäftigen, nützliche Anwendungsgebiete in ihrem Unternehmen zu identifizieren und relevante Kompetenzen im Team aufzubauen. 

Der umfassende Einfluss von KI wird sich auch auf das Anforderungsprofil an Mitarbeitende im Controlling auswirken. Qualitfikationsmerkmale wie eine hohe Affinität zu Zahlen, Erfahrungen mit Tabellenkalkulationen in Excel und Branchenwissen werden zukünftig weniger relevant sein, während Know-how für den Umgang mit BI-Tools und ein Verständnis für IT-Systeme immer wichtiger werden.  

Despite the major upheaval brought about by AI, it will not make controllers superfluous in the foreseeable future. Rather, they will be freed up for activities that are critical to value creation in the company, as they will have to spend less time collecting and processing data. In addition, AI will enable them to create more in-depth data analyses than before. As a result, new areas of responsibility will develop and there will be a shift in the work areas of controllers and machines.