LLMs are not just LLMs. The dynamic selection of large language models (LLMs) is becoming an important trend for companies that want to integrate generative AI into their processes. Given the variety of models available and their different strengths, more and more companies are turning to dynamic selection mechanisms to choose the most suitable model depending on the use case. This flexibility makes it possible to use models according to task requirements - for example, a multimodal model for complex inputs or a more specialized, smaller model for industry-specific queries. This leads to improved performance, reduced costs and a more efficient use of computing resources.
Open-source models offer high quality and innovation. The development and availability of open-source AI models promote broad use, especially by small companies, educational institutions and individuals. European initiatives such as the EURO-LLM or, more recently, OpenGPT-X’ projects are focusing on models that take cultural characteristics and European linguistic diversity into account. These show how open-source approaches contribute to the sustainable and inclusive use of AI - an important step towards technological equality and digital sovereignty.
In organizations, there is an increasing need to extract relevant information from the organization’s own data, but also to combine this with relevant ‘world knowledge’. In the corporate context, AI that uses the organization’s own data should be also able to be enriched with publicly available knowledge from the internet, for example with relevant market or industry data. Enterprise search solutions in conjunction with RAG (Retrieval Augmented Generation) are recommended to operate securely and in compliance with data protection regulations, as they enable a clear separation between internal and external data, reliably enforce access rights and ensure that external information is only retrieved and processed in a targeted manner without jeopardizing sensitive internal data or passing it on externally without authorization.
Companies and public authorities that operate chatbots in customer and citizen service see a measurable potential to increase efficiency, especially when the chatbots are enriched with relevant organizational data. In terms of language, chatbots are already at a high level thanks to generative AI, and they are also getting better and better in terms of content. While they solve simple tasks efficiently, they are still reaching their limits when it comes to complex customer enquiries. A hybrid approach of automated systems and human support therefore remains necessary to meet customer expectations.
Internal AI-supported systems can support service or helpdesk workers by analyzing incoming enquiries and comparing them with previous cases in order to suggest relevant solutions to problems. This enables employees to provide targeted support with appropriate answers or instructions that are tailored to the customer's specific concern. AI-generated answers can be enhanced with up-to-date expertise and returned to the information pool again for future use in similar cases.
The benefit for organizations: chatbots can handle most of the simpler enquiries, while employees can respond more quickly and precisely to complex requests with AI support. The number of such support systems will therefore continue to increase in 2025.
In the area of corporate search, the old search world and the new chat world will continue to converge. On the one hand, many companies still want to aggregate their data and offer their employees classic search engines for documents and, on the other hand, enable them to chat with the organization’s own data. Neither world can solve the requirements of modern information needs in organizations on its own.
Therefore, both worlds will continue to converge - on the one hand, the classic search, which is broad and exploratory and provides a lot of context around the information, and on the other hand, the new chat world, which provides precise answers to specific questions, or summaries of documents found.
In this context, intent detection plays an increasingly important role.
AI systems are continuously being optimized to better understand what the users intend with their questions and what expectations they have. This will result in new paradigms for the search, in which, for example, queries can also be made to the software to present optimal results.
After the AI hype, many companies have sobered up and realized which use cases do not work. The focus is therefore on applications that have proven their usefulness and that boost efficiency, such as the support for service employees in customer support. Such proven use cases will be professionalized and further expanded in 2025.
As long as more complex use cases do not yet work smoothly, companies will still refrain from using them in practice. We therefore recommend focusing on simple and widely applicable use cases first. These can be implemented quickly, offer clear added value and strengthen trust and expertise in dealing with AI through initial success.
Explainable AI will remain a key requirement in 2025. Users and customers expect understandable explanations of how AI obtains the presented results - a decisive factor for the widespread use of such technologies. Regulatory requirements such as the EU AI Act reinforce this trend, as they stipulate transparency and traceability. Companies must therefore ensure that the functioning of their AI systems is clear and comprehensible to both ensure compliance and gain the trust of their users.
In terms of search technology, this means that companies and authorities are rightly interested in how the hits and answers from their database are generated.
Until now, employees have used generative AI as their individual tool. They may share a prompt or individual AI-generated results with colleagues, but they don't really use the technology together as a team. We see the trend that search is evolving from a purely individual tool to a collaborative system that is seamlessly integrated into team workflows. Platforms dynamically adapt to group needs, promote real-time collaboration and facilitate the sharing and utilization of knowledge. With flexible and adaptive user interfaces, this new form of search will not only increase the productivity of teams, but also increase the efficiency and intuitivity of decision-making.
AI agents that can take on tasks independently and optimize business processes will slowly make their way into companies in 2025. Although numerous frameworks for agents already exist, they are not yet widely used in the enterprise sector. However, it is expected that 2025 will see a significant boost in the implementation of these technologies, allowing companies to benefit from more efficient workflows and increased productivity.
As agent-based systems mark a shift towards proactive AI solutions, the focus is on accountability and the ability to monitor results and processes.
