Competitive advantages through generative AI: Expertise as the key to implementation

Brüggemann, Immo; Buse, Stephan; Partuschke, Janin; Villarreal, Nohemi Working Paper Competitive advantages through generative AI: Expertise as the key to implementation Working Paper, No. 118 Provided in Cooperation with: Hamburg University of Technology (TUHH), Institute for Technology and Innovation Management Suggested Citation: Brüggemann, Immo; Buse, Stephan; Partuschke, Janin; Villarreal, Nohemi (2025) : Competitive advantages through generative AI: Expertise as the key to implementation, Working Paper, No. 118, Hamburg University of Technology (TUHH), Institute for Technology and Innovation Management (TIM), Hamburg, https://doi.org/10.15480/882.14322 This Version is available at: https://hdl.handle.net/10419/313610 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ Working Paper Competitive advantages through generative AI: expertise as the key to implementation Immo Brüggemann, Dr. Stephan Buse, Janin Partuschke and Nohemi Villarreal January 2025 Working Paper 118 Hamburg University of Technology (TUHH) Institute for Technology and Innovation Management Am Schwarzenberg-Campus 4 D-21073 Hamburg, Germany [email protected] www.tuhh.de/tim 1 This article was originally published in 2025 by the Institute for Technology and Innovation Management in cooperation with CREATUM GmbH Hamburg as part of the OpenInnoTrain Project 1 Copyright information: This work including all its parts is protected by copyright. The work is licensed under the Creative Commons Attribution 4.0 International License (CC BY 4.0, https://creativecommons.org/licenses/by/4.0/legalcode.de). Excluded from the above license are parts, images and other third-party material, if marked otherwise. DOI: https://doi.org/10.15480/882.14322 ORCID: Stephan Buse: https://orcid.org/0000-0002-4308-1707 1 This research has received funding from the European Union’s Horizon 2020 research and innovation programme, within the OpenInnoTrain project under the Marie Skłowdowska-Curie grant agreement no. 823971. The content of this publication does not reflect the official opinion of the European Union. Responsibility for the information and views expressed in the publication lies entirely with the author(s). 1 Competitive advantages through generative AI: expertise as the key to implementation Abstract The Problem In order to understand generative AI and its mechanisms of action, players need a high level of abstraction (Saitta and Zucker, 2013). However, many companies lack the necessary expertise for the technology. In addition to cultural factors, this lack of expertise is the main barrier to a resultsoriented implementation of the technology (Campos Zabala, 2023). The Solution To identify valuable fields of application, companies need to understand the expertise in the decision-making processes of their own activities. The article shows concrete examples and steps that companies can use to test and apply new technological possibilities to their own use cases. Immo Brüggemann [email protected] CREATUM GmbH Am Sandtorkai 32, 20457 Hamburg, Deutschland Dr. Stephan Buse [email protected] Institute for Technology and Innovation Management, Hamburg Universtiy of Technology (TUHH) Am Schwarzenberg-Campus 1, 21073 Hamburg, Deutschland Janin Partuschke [email protected] CREATUM GmbH Am Sandtorkai 32, 20457 Hamburg, Deutschland Nohemi Villarreal [email protected] CREATUM GmbH Am Sandtorkai 32, 20457 Hamburg, Deutschland 2 Contents 1 Introduction 3 2 New technological possibilities 3 3 Multi-agent AI and graph neural networks: revolutionizing the way we deal with complexity and knowledge 4 4 The impact of new technological opportunities on business activities 5 5 Application examples of generative AI 7 5.1 Use case I: Monitoring and assessment of machine states ..........................................7 5.2 Use case II: Multi-agent AI in marketing .....................................................................8 6 Identification of areas for the implementation of multi-agent AI in the company 9 7 Conclusion 10 Bibliography 11 Working Paper No. 117 Br ü ggemann, Buse, Partuschke and Villarreal 3 1 Introduction Many companies are discussing the potential applications of generative artificial intelligence (AI). However, in order to understand generative AI and its mechanisms of action, the actors involved need to have the appropriate technological expertise. However, many companies of all sizes lack precisely this (Campos Zabala, 2023). This article provides an overview of selected technological possibilities of generative AI, which help companies to carry out value creation activities more efficiently and effectively. However, technological expertise alone is not enough to achieve competitive advantages with generative AI. In order to identify valuable fields of application, companies need to understand the expertise in the decision-making processes of their own activities. In addition to the new technological possibilities, the article shows concrete examples and steps that companies can use to test and apply these to their own use cases. 2 New technological possibilities Technological advances in the areas of computing power (Moore, 1965), transmission speed (Butter) and memory (Kryder, 2005) follow exponential growth curves. The fact that technological possibilities are continuously and significantly improving on this basis is therefore nothing new. The problem is that humans are not able to intuitively understand exponential growth. A fact that is impressively described by the rice grain legend surrounding the creation of the chessboard in India. The impact of generative AI on the consciousness and strategic considerations of decisionmakers in all industries was correspondingly great. One observation is that the revolutionary nature of the publication of new technological capabilities is often followed by a more evolutionary phase of optimization of this technology. Agrawal, Gans and Goldfarb describe a direct consequence of this relatively silent and continuous improvement in their book “Prediction Machines” as follows: “This is simple economics: when the cost of something falls, we do more of it. [...] More significantly, because it [artificial intelligence] is becoming cheaper it is being used for problems that were not traditionally prediction problems.” Technological progress is therefore not only making the technology more capable, but also considerably cheaper. The result is that new fields of application are being opened up in which the use of artificial intelligence was previously too complex or simply too resource-intensive. (Agrawal, Gans and Goldfarb, 2018) Companies are asking themselves how they should deal with generative AI, i.e. how they should use it. Dealing with generative AI According to McKinsey (Lamarre, 2024), companies can choose and combine three basic archetypes (Maker, Taker and Shaper) to integrate generative AI into their business models: I. Maker: Makers are investing heavily in the development of their own AI technologies and platforms. A process model that is likely to exceed the resource availability of most companies many times over and is therefore primarily pursued by large technology companies such as OpenAI or Google. II. Taker: Takers use and integrate existing, widely available AI technologies into their business processes. While this offers a more cost-effective way to implement AI, it rarely enables sustainable medium-term competitive advantages to be achieved as the same technologies are also available to competitors. Working Paper No. 117 Br ü ggemann, Buse, Partuschke and Villarreal 4 III. Shaper: Shapers use basic models and modify them with their own company's data and specifications in fields of application that affect the company's core processes. This approach offers the greatest potential for companies to achieve significant competitive advantages. Before the question of how companies can use generative AI in selected value creation activities is answered in chapters 5 and 6, the technological foundations are briefly outlined in the following chapter. 3 Multi-agent AI and graph neural networks: revolutionizing the way we deal with complexity and knowledge The autosapient model of AI applications A common mistake when formulating the goal of an AI project is that AI solutions are designed with the aim of providing a final “right” solution and automating decision-making. The assumption that generative AI completely relieves decision-makers of their work and independently presents decisions that are undoubtedly correct is definitely wrong. AI is an auxiliary tool that can support managers in the decision-making process. The concept of “autosapient” AI, as described in the Harvard Business Review article “Leading in a World Where AI Wields Power of Its Own”, brings with it a different perspective on the formulation of goals for AI systems. Autosapient systems are designed to learn autonomously, continuously improve and interact with human actors. (Heimans and Timms,2024) This can take place at different levels with varying degrees of complexity. Four levels of AI models In order to better understand the transformation of AI systems, it is necessary to look at the development from simple models to complex multi-agent systems. Four levels can be distinguished, which correlate with increasing task complexity. (Guo et al. 2024; Parthasarathy et al. 2024): I. Simple models These are basic language models (LLMs) that process and generate natural language. They are suitable for general information queries and simple decision-making processes. LLMs can analyze large amounts of text and provide simple answers. However, their capabilities are limited to processing static information. II. Specifically trained models These models are tailored to specific tasks or domains and provide more detailed and contextual insights based on industry or company-specific data. The focus is on optimized output that goes beyond generic answers and includes specific expertise. III. AI agents An AI agent can perform complex, specialized tasks autonomously and interact with thirdparty systems via interfaces. It acts as an advanced decision-making assistant for a specific area of application. IV. Multi-agent AI These systems consist of several specialized agents that work together to perform highly complex tasks with a high degree of reliability. Each agent contributes its own partial expertise, resulting in comprehensive and coordinated decision-making. Each agent can be based on different technological foundations. Graph neural networks in Working Paper No. 117 Br ü ggemann, Buse, Partuschke and Villarreal 5 conjunction with graph databases, which are explained in the necessary depth below, are particularly relevant to the central idea of this article, which is to transfer company expertise into AI models. Graph neural networks and graph databases Graph neural networks (GNN) are a form of deep learning specifically designed to process complex data structures in the form of graphs. A graph consists of nodes (entities) and edges (relationships) that represent connections between the nodes. GNNs make it possible to understand and process both individual nodes and the relationships between them, which distinguishes them from conventional neural networks. The underlying graph data can come from a variety of sources, including social networks, molecules, or even graph databases specifically designed to store and query graph structures. (Khemani et al., 2024) A specific form of graph database is knowledge graphs, which accumulate domain-specific knowledge and make it explicitly accessible for AI applications. (Peng et al., 2023) By integrating knowledge graphs, especially deep information, AI models can be improved in their performance. (Elnagar and Weistroffer, 2019) Chapter 5 graphically illustrates a specific use case. How the innovative technological possibilities outlined in this chapter can be used to support the execution of value creation activities is described below. 4 The impact of new technological opportunities on business activities Businesses are driven by activities that are carried out within core, support and management processes. According to Agrawal, Gans and Goldfarb (2018), these activities follow a generic architecture, which is illustrated in Figure 1: Figure 1: Anatomy of an activity according to Agrawal, Gans, and Goldfarb (2018) At the center of the architectural model is the prediction element, in which the situation is assessed and possible outcomes are predicted. It is the basis for human judgment and subsequent action. The prediction element can be carried out by both humans and machines, explicitly by predictive AI models. The book published in 2018 describes cases as suitable for AI support if they involve large amounts of historical data, repeatable, consistent situations and clear, measurable Working Paper No. 117 Br ü ggemann, Buse, Partuschke and Villarreal 6 goals (Agrawal, Gans and Goldfarb, 2018). The availability of generative AI in the form of large language models, especially in conjunction with graph databases, expands the types of possible use cases in which the prediction module can be executed by machine. Increasingly complex, knowledge-based questions can be processed explicitly. (Elnagar and Weistroffer, 2019) A distinction is first made between experience and expertise based on Malhotra and Bazerman (2008). Experience describes the frequency with which an activity is carried out and therefore has a direct effect on efficiency, but only an indirect effect on the quality of the activity. Expertise, on the other hand, comprises the methodological component of a decision, specifically the explicit knowledge in the assessment of a situation. Expertise has a corresponding effect on the quality of the result and the effectiveness of the activity. This distinction between experience and expertise is particularly relevant when it comes to the expansion of machine-aided decision-making. In this context, two types of cases can be identified, which are explained in more detail in the following chapter using specific application examples. I. Cases in which the expertise is not explicitly available in the company This expertise can be developed using generative AI on the basis of unstructured internal and external data. To do this, the unstructured data is converted into a knowledge graph using a language model, which is then used in the predictive model and improves the result (Elnagar and Weistroffer, 2019). Specific sources can be processes from a customer support ticket system, machine manuals or email correspondence. II. Cases in which there are no measurable results of the activities In activities that do not focus on classic predictive issues, the integration of expertise in language models can be used to support decision-making. Specifically, for example, expertise on strategic management methods can be integrated into autosapient AI assistants that support decision-makers in strategy development and implementation. (Csaszar et al., 2024) The consequence for companies Companies must understand these new opportunities and reflect them on their activities in order to develop competitive advantages and new forms of value creation by shaping the technology (see chapter 2). Two key questions arise explicitly: I. How great is the potential for expertise: What expert knowledge is available in the company in an unstructured or incomplete form and can be made accessible and integrated in a scalable way using forms of AI? II. How can the new technological possibilities be used to convert existing expertise into new forms of value creation? The following chapter describes two examples of use cases that illustrate the advantages of the innovative process approach.