Skip to main content

Unlocking the Business Potential of Large Language Models: Real-world Applications and Obstacles

by Jochen Wulf and Jürg Meierhofer – (Data-Driven Service Engineering Group at ZHAW)

At this year’s IEEE Swiss Conference on Data Science there was a very informative workshop on Generative AI in Practice. The presentations and discussions in this workshop made clear that the generative AI technologies, and Large Language Models (LLMs) in particular, are very versatile and powerful. It also became apparent, however, that the business potential of LLMs largely remains unclear.

Figure 1: DALL-E Visualization of a Talking Machine

OpenAI’s AI chatbot ChatGPT has already gained over 100 million users within the first two months of its release. This makes this internet service the fastest growing of its kind. In comparison, the runner-up, Tiktok, took a full nine months to reach a similar number of users. The potential of the underlying technology, LLMs, is undisputedly recognized by business leaders. According to a study by Gartner among business executives1, 45% of respondents have already intensified their investment in artificial intelligence (AI) as a result of the success of ChatGPT.


LLMs are extremely large-scale artificial neural networks that are trained with terabytes of textual content to complete texts. LLMs can therefore generate new content and thus belong to the class of generative AI solutions. In contrast, discriminatory AI models can only establish assignments or classifications between different inputs and predefined outputs.

LLMs can be used for a variety of purposes, such as text summaries, sentiment analysis, or named entity recognition. Although there are already initial indications of the productivity gains that can be achieved through the use of LLMs, the role of this technology for the business models of industrial companies is still largely unclear.

The Impact of LLMs on Business Models

In the following, we present the findings derived from own prototypes and a comprehensive analysis of more than 50 real-world use cases pertaining to the application of LLMs within various companies. A distinction can be made between four mechanisms of how business models are changed with LLMs:

  • new customer benefits
  • new sales and communication channels
  • increased business process automation
  • improved use of information resources

New Customer Benefits

LLMs play a crucial role in operating personal assistance systems. Instacart, a grocery delivery service, utilizes LLMs to address nutrition queries and offer personalized product recommendations.

Furthermore, LLMs serve as personal coaches, particularly valuable in the realm of learning. Khan Academy, an educational platform, employs LLMs to detect errors in programming tasks and generate helpful solution hints.

LLMs also possess the capability to independently generate content that is relevant to customers., an online service, exemplifies this by creating blog posts, social media content, and website material based on bullet-style keywords and predefined language styles.

Additionally, LLMs facilitate voice-based interactions with machines. Mercedes, for instance, integrates LLMs into the infotainment systems of their premium vehicles to provide comprehensive answers to complex customer questions while driving.

New Sales and Communication Channels

LLM-based chatbots offer significant advantages in automating sales and customer service processes. In Switzerland, the insurer Helvetia has successfully implemented an LLM-based chatbot to handle inquiries regarding their product range.

Another notable example is Solana, a blockchain operator, leveraging ChatGPT in their customer service operations. By utilizing LLM-based chatbots, Solana effectively assists customers in resolving intricate service-related challenges, ensuring a seamless user experience.

Increased Business Process Automation

LLMs offer remarkable potential in enhancing automation within information-intensive business processes. The Radisson hotel chain has effectively employed LLMs to automate the handling of customer inquiries and cancellations, enabling swift and accurate responses. Additionally, LLMs generate helpful suggestions for emails and review responses, streamlining communication and enhancing customer satisfaction.

Another notable application is observed in Swiss Migros Bank, where LLMs play a pivotal role in partially automating mortgage application processing. By intelligently recognizing case-specific requirements and evaluating text-based customer documents, LLMs assist in expediting and improving the efficiency of the application review process.

Improved Use of Information Ressources

The fourth mechanism focuses on the enhanced exploitation of information resources. Morgan Stanley, a securities trading company, exemplifies this by leveraging LLMs to facilitate employee access to and evaluation of internal documents. Through the application of LLMs, Morgan Stanley streamlines the process of retrieving and analyzing crucial information, ensuring efficiency and informed decision-making within the organization.

Likewise, Zurich Insurance capitalizes on LLMs to automate contract evaluation and ascertain the validity of insurance claims. This strategic employment of LLMs empowers Zurich Insurance to effectively and efficiently assess the presence of a claim, ultimately leading to enhanced operational processes.

Figure 2: Four Mechanisms in Osterwalder´s Business Model Canvas

Current Challenges in the Commercial Application of LLMs

When evaluating the strategic importance and necessity for action concerning the commercial application of LLMs, companies are faced with three fundamental questions.

What are technical risks associated with the use of LLMs?

One significant challenge arises from the risk of LLMs generating false or inaccurate statements, commonly known as hallucinations. However, advancements in prompt engineering, which involve carefully formulating textual instructions, have already proven effective in mitigating this risk to a considerable extent. Additionally, the development of fact-checking methods is underway to ensure that the output generated by LLMs is rooted in accurate and verified information.

Another crucial technical concern revolves around the security of sensitive data shared during the prompting process. The potential exists for malicious actors to employ targeted prompts, referred to as Training Data Extraction Attacks, to extract training data from LLMs. Consequently, it is imperative to eliminate the possibility of shared data being utilized to train publicly accessible LLMs. Alternatively, dedicated LLMs can be utilized to safeguard the confidentiality of shared data.

What legal framework conditions need to be taken into account?

When utilizing LLMs, it is essential to adhere to relevant data protection regulations, particularly if personal data is being processed. This entails fulfilling information obligations and respecting individuals’ rights to information, similar to other AI applications.

Additionally, companies need to consider the evolving legal landscape, such as the European Union’s AI Act. The current draft of the AI Act specifies certain requirements for LLMs, including the prevention of generating illegal or discriminatory content and the disclosure of copyrighted content used during training. However, a comparison of different LLMs reveals that most models do not fully comply with these requirements, particularly regarding copyright compliance. Therefore, it is crucial for companies to carefully assess and ensure compliance relevant legal obligations when making long-term technology decisions involving LLMs.

In which areas should companies invest in LLMs?

Providers of standard software and Internet services are already investing heavily in LLMs. This includes areas such as sales management, customer service, marketing and knowledge management. Non-software companies will likely source such software rather than build it themselves.

More interesting for non-software companies are application areas in which LLMs have a direct impact on their value propositions or business-critical business processes. For example, robotics manufacturer Boston Dynamics uses LLMs to enable voice-based interaction between users and machines. Ivaldi, a distributed production specialist, uses LLMs to help maintenance teams troubleshoot. Rolls-Royce uses artificial intelligence to harness unstructured data and optimize supply chain management.

These illustrations highlight the substantial innovation potential of LLMs, extending beyond software companies to various other industries. Particularly noteworthy is the possibility for non-software companies to harness this potential by reimagining user interactions or unlocking significant optimization opportunities.