Companies are increasingly recognising the potential of AI for internal applications. The challenge, however, is to select the most appropriate use cases from the many possible new ones. It is therefore advisable to first identify and successfully implement a particularly relevant use case. The experience and results from this lighthouse project can then serve as a guideline for the effective implementation of further use cases in the company.
In this guide, we will show you step-by-step how to select the optimal AI use case for your company, with a particular focus on use cases for medium-sized manufacturing companies.
Companies are increasingly recognising the potential of AI for internal applications. The challenge, however, is to select the most appropriate use cases from the many possible new ones. It is therefore advisable to first identify and successfully implement a particularly relevant use case. The experience and results from this lighthouse project can then serve as a guideline for the effective implementation of further use cases in the company.
In this guide, we will show you step-by-step how to select the optimal AI use case for your company, with a particular focus on use cases for medium-sized manufacturing companies.
Every week, new ways in which generative AI can be used in companies are emerging. Most use cases fall into six main categories:
By using AI, material consumption and waste can be reduced and optimized processes implemented, among other things.
Employees are equipped with the right tools to efficiently complete repetitive tasks with the help of AI.
By using AI to manage knowledge databases, companies can systematically record, organize and make available the know-how and experience of their employees.
AI and sensors enable seamless monitoring and documentation in production and logistics to optimize processes for efficient manufacturing.
AI is used to support the development and optimization of innovative products and designs, to improve details or to implement new solutions.
Better decisions are made possible by data-based forecasts and insights from business figures and process monitoring.
The effectiveness of these applications increases with the performance of AI models, in particular their ability to understand complex relationships. More detailed explanations of use cases can be found here.
The right AI use case is identified in four focused steps:
The first step is to identify the focus area within the business that you want to focus on first. This could be a specific business unit, location or process within the value chain. It is advisable to choose areas where there is already a high level of acceptance among employees and/or where you already have a high level of expertise. This will allow you to identify potential more quickly and efficiently.
Based on lean management principles, the first step is to analyse the current state of the selected process. Various process techniques, such as value stream mapping, can be used to do this. Next, an estimate is made of (1) how much time is required for each process step, (2) what costs are associated with that step, and a first indication of (3) how much data is already collected in each process step (high / medium / low). The areas of high time, high cost and high data are best suited for in-depth analysis in the next step.
Once a relevant use case has been identified, the next step is to find the optimal AI solution for it. Interdisciplinary innovation workshops are particularly suitable for this. Prior to these workshops, it makes sense to analyse certain questions in advance, e.g. whether an in-house solution should be developed (build) or whether a suitable existing solution should be purchased (buy). It is worth taking a look at existing internal capabilities on the one hand and existing solutions on the market on the other. When searching for existing solutions, it is also worth looking beyond industry boundaries.
The previous findings can then be collected and processed in an interdisciplinary ideas workshop. We recommend conducting such innovation workshops in interdisciplinary teams that are mixed with (1) people with AI knowledge to keep an eye on the limits of what is feasible, (2) people who are involved in the process step themselves to assess actual user needs, (3) people with decision-making authority and (4) external parties who can take an undistorted look at the whole thing. During the idea workshop, with the help of various creativity techniques company-specific use cases are generated and classified on opportunity radar (see opportunity radar).
The final use case can be prioritised based on business decision factors. An impact-effort matrix can be used for a holistic decision. In a second step, the relevant use cases can be further subdivided by specifying the "effort" in terms of (A) time and (B) complexity and the "impact" in terms of (A) monetary and (B) not (directly) monetary impacts (e.g. customer satisfaction, reduced decision time, etc.).
This methodical approach makes it possible to systematically identify and evaluate AI use cases to ensure that they generate the greatest value for the business.
The most commonly used KPI for projects is ROI (return on investment). In order to determine the exact ROI, the monetary turnover of the solution must be evident, i.e. the AI solution must be monetisable.
In the case of customer-facing AI use cases, monetisation often occurs either directly through a premium or indirectly through increased order volume as a result of the improved value proposition.
For use cases that relate to internal business processes, monetisation is generally less straightforward. Value is usually created indirectly through cost savings, efficiency gains, quality improvements, etc. A before and after comparison is therefore particularly valuable for internal projects. For this purpose, a success relevant KPI is selected and measured both before and after the implementation of the AI system. Possible examples are
As with any experimental test, care must be taken to ensure that no confounding variables distort the results. For example, the error rate in make-to-order production may increase in the short term because production workers put extra effort into the artificial test environment - a phenomenon known as the experimenter expectation effect. Therefore, it is generally useful to choose a long-term time horizon.
When using AI in a business context, companies will need to comply with the requirements of the EU AI Act, which will come into force in 2026. There are two key aspects to this:
AI applications must be categorised into four different risk levels and secured accordingly: Unacceptable Risk (e.g. social scoring, facial recognition), High Risk (e.g. recruitment or lending systems), Limited Risk (e.g. chatbots) and Minimal Risk (e.g. spam filters). Each risk category requires specific measures to ensure that the solutions meet the relevant requirements and comply with legal regulations.
Article 4 of the EU AI law requires the implementation of training programmes tailored to the specific needs of the company and the existing technical skills of its employees.
Our article on EU AI Act outlines the key things SMEs need to know about the EU AI Act.
