Get answers to internal processes, resolve helpdesk cases faster, relieve employees of recurring tasks, get AI support for creating reports, and much more. Companies and public authorities have numerous ideas about where AI could help them work more efficiently. But how exactly should they proceed? Here are the steps:
‘We need AI now’ as a management directive is a very poor starting point for introducing generative AI. Before starting a project, consider the following: What specific problem do you want to solve? Is it about saving time when researching information, protecting against loss of expertise, or improving customer communication? The more clearly you define the use case, the more precisely you can derive and prioritize requirements.
Practical tip: Analyze existing processes and identify the greatest potential for efficiency gains. This will give you a solid foundation for all further steps.
Even the best use case will fail if it fails to meet the needs of users. Therefore, involve the people who will later be working with the solution from the outset, as well as those in your organization who need to be formally involved, such as the works council, data protection and HR departments. Not everyone is enthusiastic about AI, so employees should be made aware of it, involved in the process and their fears taken seriously. The project should be seen as the beginning of a necessary cultural change in the company, which involves not only technical and professional implementation but also internal communication and coordination tasks.
Practical tip: Clarify what these user groups expect and identify their biggest pain points. What takes up the most time today? Start a small survey or workshops to gather requirements and wishes. This will ensure acceptance and avoid surprises later on.
A project can only be successful if it is clear in advance what success means. Define how you will measure the success of the project. Are there KPIs such as reduced search times, more queries answered per day or higher user satisfaction?
Practical tip: Also define clear ‘showstoppers’: for example, that the AI does not hallucinate and communicates clearly when there is no relevant information, or that the solution is GDPR-compliant. This creates transparency – both internally and towards your implementation partner.
A proof of concept (PoC) is advisable – i.e. a method of checking feasibility before investing further resources. However, a PoC only makes sense if it reflects real scenarios and allows for evaluation of expected benefits.
Practical tip: Prepare specific use cases to be tested in the PoC, e.g. typical support queries that a chatbot should answer. Which cases must be fulfilled, and which would be nice to have?
With internal search solutions and chatbots, not every employee should be able to see all the information. Therefore, the following should be clarified from the outset:
Practical tip: Clarify internally which data may be processed, which protective measures are necessary and how rights and role concepts can be implemented technically. This will help you avoid delays in the project and build trust among stakeholders.
Consider the infrastructure as well. Where will the system run later? On-premises – i.e., on your own data infrastructure – for maximum data control, in the cloud for greater scalability, or as a hybrid solution?
Seek advice on which large language model (LLM) best suits your use case. Not every model is suitable for every industry. Data protection, integration capability, and costs play a major role here.
Practical Tip: You can find more information on this in our blog ‘Guide to choosing the right LLM”.
After the PoC, the project begins, which is only completed with a successful rollout. Plan sufficient resources for training, support and internal communication at an early stage so that everyone involved understands and actively uses the new solution.
Practical tip: Think about future expansion now. AI can assist not only with search, but also with generating email responses, translating texts, and analyzing image content. Proceed step by step and expand the possibilities in a targeted manner.
An AI project is neither a sprint nor a marathon, but rather a well-planned journey. With a clear vision, the right questions and a strong partner, it will be a lasting success. Whether you want to chat with your organization’s data or directly with the LLM – with our digital assistants iAssistant and iHub, both are smart and secure
Would you like to know what a PoC could look like in your company? We will accompany you every step of the way.
