Ever since the hype around generative AI began, companies have been grappling with how and where to apply this technology. Pioneering projects such as dmGPT from dm have already provided an insight into the possibilities. However, it remains unclear which use cases are actually suitable, how they can be implemented effectively and how employees can be empowered. This article provides a clear overview of existing AI systems and the associated risks. The focus is on Enterprise GenAI solutions. Other interesting use cases can be found under Use Cases.
Ever since the hype around generative AI began, companies have been grappling with how and where to apply this technology. Pioneering projects such as dmGPT from dm have already provided an insight into the possibilities. However, it remains unclear which use cases are actually suitable, how they can be implemented effectively and how employees can be empowered. This article provides a clear overview of existing AI systems and the associated risks. The focus is on Enterprise GenAI solutions. Other interesting use cases can be found under Use Cases.
This category includes third-party software solutions that are used without deep integration with internal systems. An example is ChatGPT Plus. Such tools are ideal for specific functions and individual users, offer a relatively low cost of entry and low maintenance costs, but have limited customisation capabilities.
These are purpose-built applications for internal use only, such as an expert chat system. These applications allow complete control over internal data, effective user management and can be fully integrated with existing systems. They also offer extensive customisation options.
These applications are aimed at external users and play an important role in customer interaction, such as a self-help application. They contribute to the brand image, improve the user experience, offer value-added services to customers and must meet high standards in areas such as data protection (GDPR) and reliability.
The use of generative AI models in organisations brings with it four risk factors that need to be carefully assessed to ensure that the use of these technologies is effective and compliant:
Improper handling of data, particularly in customer-facing applications, can lead to the inadvertent disclosure of sensitive or proprietary information. This underlines the need for strict data cleansing and clear terms of use.
Attackers can overload LLMs with resource-intensive requests, resulting in a degradation of service quality or increased operational costs. Self-hosted LLMs are particularly vulnerable due to the infrastructure required.
The use of third-party LLMs that charge by usage can lead to unexpectedly high costs without proper monitoring and validation. Attacks that artificially inflate usage volumes can also lead to financial losses.
Over-reliance on LLM spend can lead to poor decisions and legal issues, especially if the models produce false or misleading information, a phenomenon known as 'hallucination'.
Retrieval Augmented Generation (RAG) is a technique that improves the accuracy of generative AI models by integrating external information sources. RAG combines AI model generation processes, in particular Large Language Models (LLMs), with external databases or resources to retrieve relevant facts on current or specific topics.
This method reduces the risk of incorrect or irrelevant answers (‘hallucinations’). By adding verifiable data sources, the models can provide more accurate and reliable information, which increases reliability and improves the user experience.
In order to effectively empower employees to use generative AI, such as Retrieval Augmented Generation (RAG), the following measures should be implemented:
Targeted training on the specific risks of GenAI is essential. This should include practical examples to teach the correct use of the technologies, in particular to minimise the risk of misinformation.
Policies should allow employees to experiment with GenAI. It is important that management supports these efforts and recognises the importance of innovation for technological development and competitiveness.
Regular workshops and webinars to share experiences and challenges are crucial. These formats support not only knowledge sharing, but also collective problem solving and innovation within the organisation.
These approaches will ensure that teams are well prepared to effectively address the challenges and opportunities of GenAI. Empowering teams will become even more important in light of the EU's AI Act, which will come into force in 2026, and Article 4 of which stipulates, among other things, that employees who come into contact with AI systems must be trained accordingly.
