Methods and models for the identification and evaluation of AI use cases

Abstract

Artificial intelligence (AI) is considered to be one of the most important future technologies and is increasingly applied by both private and public organizations. As a multi-purpose data-driven technology, likely any organization may profit from AI’s wide range of applications. However, the majority has not adopted AI yet and several initiatives to do so fail. This is especially true for organizations from the non-tech sectors but also small and medium-sized organizations. To adopt AI, several readiness factors should be present: IT, data and financial resources must be provided and aligned to AI’s requirements while an innovative company culture is also beneficial. The strategic alignment of an AI solution is required as well, i.e., the AI use cases must fit with strategic and operative business goals as well as existing operations. Identifying purposeful AI use cases that fulfil those criteria already poses a severe hurdle for several organizations. This dissertation aims to support such organizations by providing procedural guidance and practical recommendations that help to identify and evaluate meaningful AI use cases. A procedure model providing detailed activities, instructions, techniques, tools and roles is proposed alongside an evaluation framework and modeling techniques supporting the identification and evaluation of AI use cases. The basic idea is to identify a set of AI use cases that are step by step evaluated and specified. To find business-driven ideas for AI use cases first, two strategies are proposed. A top-down approach that analyzes strategic business goals down to operative goals and associated processes is suitable to find ideas for AI use cases that promise short-term improvements for existing business solutions. A user-centered bottom-up approach instead focuses on the challenges and whole working environment of selected employees and derives AI enabled solutions to support them. This leads to ideas for AI use cases that are disruptive and offer great business improvements but are not feasible on the short to medium term. The feasibility of each AI use case idea is mainly explained by the availability and quality of required data as well as the insights when exploring the data. The AI use case ideas from the user-centered bottom-up approach are often not supported with required data. Thus, a lot of prior work is required to enable the use case goals. The top-down approach in contrast leads to AI use cases that are well aligned with existing data and thus promise a higher feasibility. Each AI use case should eventually be evaluated for its expected business impact against its feasibility before approaching the proof of concept phase. Here, an AI model is designed and evaluated for the real business improvement it would provide with currently available data. In addition, the AI model is tested against AI specific criteria such as explainability, transparency, legal and ethical aspects as well as the required integration into existing IT, data and human workflows. These findings are derived from a real-world case study where the procedure model was executed and evaluated within an organization. Nine business user groups were part of the project that aimed to identify and evaluate AI’s potentials. The case study revealed the complexities and issues a non-tech organization faces when trying to introduce advanced data-driven technologies such as AI.