Many companies plan to integrate artificial intelligence into their business model. Motivated and full of optimism, they simply get started - but what often follows is disillusionment: from failing prototypes to real disappointment over unforeseen hurdles. Here it quickly becomes clear that a structured, iterative approach and planning is needed. Companies must identify their use cases in advance. But how do you find them and how can you prioritize them? appliedAI have created an informative white paper on these requirements. Co-author Philipp Hartmann, Director of AI Strategy at appliedAI, gives us important insights in an interview.
Why is it difficult for many companies to find use cases for artificial intelligence?
A major problem is a lack of knowledge about artificial intelligence: On the one hand, there are often false expectations about what artificial intelligence can and cannot do. On the other hand, many companies also lack the understanding of the specifics that need to be considered when developing AI use cases - many equate AI with "digital", but many challenges are fundamentally different. This lack of understanding - by the way, especially on the executive level - leads to the fact that many companies either do not deal with AI at all or implement more or less random AI applications.
How can a company identify realistic use cases that really benefit them? Is AI useful for every company at all?
In addition to the described basic understanding of AI as a technology, the essential prerequisite for finding the "right" use cases is first of all to understand in which areas AI can create the most value for a company - this is what we call AI vision. What is meant by this can be clearly illustrated using the example of the Linde company: Linde essentially breaks down air into its individual gases and sells them to various end customers. AI does not change this for the time being. However, there are two main cost blocks: The energy for air separation and the costs for logistics. These are precisely the areas for which AI creates the most value at Linde.
But it is also clear that AI is always just a tool: depending on the application, it can be very valuable or there may be other solutions without AI. It is important to understand when this is the case.
What typical mistakes do companies make when implementing AI?
AI solutions are often developed detached from concrete problems or AI solutions are used because they are supposedly "en vogue" - a typical example are chatbots. The key to solving this problem is the cooperation between users (e.g. the business department) and developers (e.g. the machine learning engineers). Neither AI use cases, which are conceived in the ivory tower of Data Scientists, nor use cases, which are "only" developed on slides, usually lead to a good result.
How can an effective implementation of Artificial Intelligence succeed?
Companies should follow a structured process to find the appropriate use cases. Typically it should follow these four steps:
1. Preparation: Before developing AI use cases, it should be ensured - as described above - that employees have a basic understanding of AI and an AI vision that provides rough search fields. In addition, relevant use cases from industry or related fields should be collected so that there is a clear understanding of what is already available.
2. Ideation: Now the actual development of ideas for use cases begins. Here, too, a systematic approach should be taken. Starting points should always be problems or opportunities for improvement in the application fields defined in the AI vision - for example, along a "customer journey" or process maps.
3. Evaluation: In a next step, the collected use cases must be evaluated. Unfortunately, this evaluation is often only done "on instinct". We have developed guiding questions that help to systematize this step. On the one hand, you need to understand the value of the use case:
- What financial value does the use case create, through efficiencies or additional business?
- What strategic added value does the solution create?
On the other hand, you have to estimate the costs and effort of the solution:
- Do I have the required data in sufficient quality and quantity?
- Do I build on existing, known algorithms/solutions or do I have to develop them from scratch?
- Which systems and processes have to be adapted?
- Do we have the necessary technical competence as well as the expertise to implement the use case?
4. Prioritization: Based on the evaluation, the use cases can be prioritized. Of course, one would like to implement use cases first, which have a high value and are relatively easy to implement - unfortunately these are rare. Accordingly, one has to consider whether more complex use cases can be broken down into simpler ones. It is important to understand that the prioritization of use cases can never be done independently: Use cases that use the same data or require similar algorithms are usually easier to implement. This means that prioritization must always be carried out in relation to the other use cases.
Which support does appliedAI offer here?
appliedAI supports companies along the whole chain: From the development of a comprehensive AI vision to find the appropriate search fields, to AI workshops to identify suitable use cases together with the machine learning engineers and AI strategists, to the implementation of first proof of concepts (PoC).
Furthermore, we support companies in developing and implementing the necessary processes for the systematic development and implementation of AI Use Cases.
We have also compiled our experiences and recommendations on the topic of "Use Case" development in a report that you can find here.
Thank you very much for the interview!