The systematic literature review is a fundamental method in academic research, aimed at collecting, systematically evaluating and synthesizing comprehensive and objective information on a specific topic. Its goal is to generate well-founded and traceable insights that can serve as a basis for further research or practical applications.

In addition to providing a comprehensive overview of the topic, this method also enables the identification of research gaps and supports the formulation of new hypotheses, which may serve as a starting point for future studies [11] [12] .

To ensure a thorough examination of the scientific literature, the SALSA model is applied. SALSA stands for Search, Appraisal, Synthesis, and Analysis, and offers a clear methodology that supports researchers in conducting a systematic and transparent review process.

In the first step, the search phase involves the targeted search for relevant literature sources in various databases. In the appraisal phase, the identified studies are evaluated to determine their quality and relevance. The synthesis phase includes the integration of results from the studies, while the analysis phase allows for a deeper examination of the synthesized information to generate new insights [13] [14] . Figure 2 illustrates the SALSA process.

The first step in the SALSA process is the precise formulation of the research question, which was defined in Chapter 1.1 and establishes the framework for this study. Based on this, the databases to be searched and the search strategy are determined [15] . The search strategy and the appraisal process are discussed in the following chapter.

This study is based on a search in the bibliographic database Scopus. The search process was conducted from October 31, 2024, to November 3, 2024. Since English is the predominant language in academic research, only English-language literature was included in the search. The following exclusion criteria were applied: Limit To: Language: English. No further restrictions were applied. The inclusion criteria have been carefully defined to encompass studies that focused explicitly on the integration of AI within BA platforms in SMEs. Specifically, the research articles must have been published in peer-reviewed journals and written in English. Only studies published within the last ten years were included to capture the most current trends and insights relevant to AI integration in SMEs. The exclusion criteria eliminated studies that did not directly address AI applications within SMEs or those that addressed broader aspects of AI without focusing on the integration challenges. Additionally, articles that were opinion pieces, non-empirical studies or lacked primary data were not included as they do not provide robust evidence to support the research objectives.

Following the preliminary identification of articles through database searches, an initial bibliometric sorting was conducted. This involved a filtering of the results based on the set inclusion and exclusion criteria as defined in the earlier phases of the review. Articles were retained if they focused specifically on the integration of AI BA in SMEs, were published in peer-reviewed journals and were accessible in full-text form [16] [17] .

After the bibliometric sorting, a full-text screening was systematically employed to ensure that each selected article adhered to the established criteria comprehensively. During this stage, articles were evaluated in detail for their alignment with the research question posed in the study. Full texts were evaluated for methodological soundness, relevance to AI integration in SMEs and the original data they presented, as well as their contributions to understanding the challenges and benefits associated with this integration.

Concurrent with the full-text screening process, a quality grading was applied to the selected studies. This involved an independent assessment of each article which focused on aspects such as research design, data collection methodologies and generalizability of the results. Following established protocols, studies that demonstrated adequate quality (for instance: randomized control trials, systematic reviews or those with robust qualitative methodologies) were favored for inclusion in the final synthesis, reflecting a commitment to incorporating the most valid findings available [18] .

Ultimately, after a review of eligible studies, a final selection of articles was compiled for inclusion in the systematic review. This selection was driven by the insights and data provided by the articles which concerned the challenges faced by SMEs during AI integration.

The following search algorithm has been used to ensure an efficient and precise search within the article titles, abstracts, and keywords:

“challenges”AND “ai” AND “integration”

Initially, 5566 articles were found. An analysis of the search results reveals, as shown in Figure 3 , that there have been publications on the topic in recent decades, but the intensity of the topic has increased since 2018, with the peak in publications occurring in 2024. This demonstrates the intensity of research activities in this area, which can be attributed to the continuous development and relevance of the topic in practice.

An analysis of the distribution by document types shows, as illustrated in Figure 4 , that the largest share consists of scientific articles, which also indicates intensive research activity in this area. A nearly equally large share consists of conference papers, suggesting the relevance, timeliness, and achievements in this field. Reviews are the third-largest category, indicating that a significant number of review articles exist, analyzing and summarizing existing research findings.

During the appraisal phase, each identified study underwent a quality assessment to employ a standardized evaluation framework. This framework includes criteria such as methodological soundness and the relevance to the defined research questions. Each study was categorized based on its study type (qualitative, quantitative, or mixed-methods) to facilitate an analysis of the robustness of evidence provided. To refine the search and improve the relevance of the search results, a revised search algorithm was created. The previously used parameters were maintained: Limit To: Language: English. No further restrictions were applied. To ensure an even more precise search within article titles, abstracts, and keywords and to narrow down the relevance of the topic, the following search algorithm was used:

“challenges” W/500 “ai AND integration” AND “smes”

The proximity operator W/500 searches for articles where the term “Challenges” appears within a distance of 500 words from “AI Integration”. Additionally, the articles must include the term “SMEs”. This ensures that “Challenges” and “AI Integration” are thematically and linguistically closely related, while still being flexible enough to allow for variations and sentence structures. Although the length of abstracts in academic articles can vary, there are some generally accepted standards that are widely used within the academic community. The RECORD statement, which deals with the reporting of observational studies, emphasizes that abstracts should contain a concise summary of the key results and conclusions to effectively inform the readership [19] . Typically, abstracts range between 150 and 250 words, although the exact length often depends on the specific requirements of the journal and may even extend to 500 words. Therefore, the proximity operator W/500 was chosen.

The new search algorithm resulted in 31 articles found on Scopus. An analysis of the search results reveals, as shown in Figure 5 , that publications on the topic have been present since 2018, with the intensity of the topic increasing from 2022 and reaching its peak in 2024.

An analysis of the distribution by document types shows, as in the first search, that the largest share consists of scientific articles, indicating intensive research activity in this area, followed by conference papers, books, and conference reviews, as shown in Figure 6 .

As part of the identification process, a total of 5,566 articles were initially identified in the Scopus database. After refining the search algorithm, 5,535 articles were excluded, leaving 31 articles that are addressed in this study. The final step of the SALSA methodology is the synthesis process. In this phase, the included articles are examined and analyzed for their interrelationships [15] . The synthesis is discussed in the following section, the qualitative analysis.

A central aspect addressed in several studies is the need for training programs to promote AI skills in SMEs. Sandoval-Gómez emphasizes that the development of technological innovations and training programs is crucial to enhance productivity and efficiency in public institutions [20] . This research shows that the implementation of such programs can lead to improvements in employee skills, as well as a social and ecological impact. Orth adds to this perspective by highlighting the importance of AI-supported further education concepts for SMEs, which are strategically aligned with the transformation goals of companies [21] . Both studies underline that targeted training in AI technologies is essential for the competitiveness of SMEs.

The challenges associated with implementing AI in SMEs are thoroughly examined by Govori and Sejdija as well as Ridho. Govori and Sejdija identify high costs and technical requirements as the greatest obstacles for SMEs in integrating AI [22] . These findings align with Ridho’s results, who stresses the relevance of future empirical studies to explore the long-term effects of conversational AI in SMEs [23] . Both studies show that despite the potential benefits of AI, such as increased efficiency and improved decision-making, significant challenges must be overcome.

Another important aspect discussed in the literature is the role of AI in production optimization. Kangru et al. develop a concept for evaluating the performance of robot-assisted manufacturing cells, considering technical and functional capabilities [24] . This study highlights the importance of key performance indicators (KPIs) for assessing the efficiency of production cells. In contrast, Ko et al. focus on the challenges associated with implementing smart factories [25] . These different emphases show that AI integration is not limited to production but also has significant impacts on other business areas. The challenges of data collection and data management for AI applications in SMEs are addressed in the work by Steuer et al. [26] . This study highlights the difficulties SMEs face when implementing AI due to insufficient data infrastructure. This contrasts with the optimistic perspectives outlined in the works of Parkinson et al. who discuss the role of AI in strengthening SME resilience during the COVID-19 pandemic and the integration of AI into energy management systems [27] .

These differing perspectives emphasize that while some studies focus on the challenges, others highlight the positive effects of AI. Another significant point is the discussion of the ethical implications of AI usage in SMEs. Han develops an explainable AI tool to assess operational risks of AI systems for SMEs. This work is particularly relevant as it stresses the need to integrate ethical considerations into the AI implementation process [28] . The consideration of ethical aspects is crucial for fostering trust in AI technologies and ensuring that these technologies are used responsibly. The works of Zairis and Zairis as well as Agrawal et al. extend the discussion to the impact of AI on the labor market. Zairis and Zairis examine the challenges of digital innovation in a rapidly changing environment [29] , while Agrawal et al. analyze the ambivalent effects of automation on the labor market [30] . These studies emphasize that the integration of AI has not only technological but also social and economic implications that need to be carefully considered.

In addition to the commonalities between the authors, differences are also evident. A central difference between the authors lies in their focus on specific aspects of AI integration. While Orth focuses on the development of AI-supported further education concepts strategically aligned with SMEs’ transformation goals [22] , Ridho examines the opportunities and challenges of conversational AI in SMEs [24] . Orth argues that targeted education in AI technologies is essential for the competitiveness of SMEs, while Ridho emphasizes the need for future empirical studies to examine the long-term effects of conversational AI [22] . These different focuses show that the authors address different facets of AI integration, leading to different conclusions about the necessity of training programs and empirical studies.

Another significant difference lies in the perception of the challenges associated with AI implementation in SMEs. Govori and Sejdija [22] identify high costs and demanding technical requirements as the primary challenges for SMEs in integrating AI. This view is supported by Steuer et al., who examine the difficulties in data collection and management for AI applications in SMEs [27] . In contrast, Parkinson et al. emphasize that AI can be used as a tool to strengthen SME resilience during the COVID-19 pandemic [28] .

The ethical implications of AI usage are also treated differently. Han develops an explainable AI tool to assess operational risks of AI systems for SMEs, highlighting the need to integrate ethical considerations into the AI implementation process [29] . In contrast, Agrawal et al. focus on the impacts of automation and AI on the labor market, discussing the ambivalent effects of automation on various professions [30] . These differing approaches to the ethical consideration of AI integration show that there are varying views in the literature on how ethical considerations should be integrated into the implementation process. Additionally, there are differences in methodology and the approaches used. Kangru et al. employ qualitative and quantitative methods to assess the performance of robot-assisted manufacturing cells [25] , while Baiyere et al. focus on digital transformation and the new logics of business process management [31] . These different methodological approaches lead to varying insights into the effects of AI on the efficiency and effectiveness of business processes in SMEs.