Craft Expertise and Selective AI Use: A Single-Case Study of an Artisan Enterprise in Northern Thailand

Introducing artificial intelligence (AI) into an artisan enterprise raises questions about which activities it can support and where practitioner judgment remains necessary. Research on human–AI collaboration explains how human knowledge and AI capabilities can contribute to organizational decisions [1], while automation and augmentation remain interdependent [2]. Complementarity depends on how their different information and capabilities are brought together [3]. In a craft business, these questions concern both the creation of objects and how their meaning is communicated to customers. Understanding adoption therefore requires attention to decisions about particular tasks and outputs. Entrepreneurial adoption research considers domain experience and technological familiarity [4] and SME research identifies technological, organizational and environmental conditions affecting use [5]. Studies of entrepreneurial decisions and expert work also examine the role of prior knowledge in working with AI [6, 7]. Building on these perspectives, this study asks: How do craft expertise and business experience inform the selection and evaluation of AI-supported tasks in an artisan enterprise? The focus is how practitioners apply their knowledge during implementation, including the preparation of business information and assessment of generated outputs. Gung Mog Kaam Crafts & Arts, a ceramics enterprise in Pai, northern Thailand, provides a case in which craft expertise, technological familiarity and initial AI implementation can be examined together. Its owners value their creative contribution and introduced AI into marketing and knowledge communication while retaining ceramic design. Access through a consulting engagement provided field notes, implementation records and subsequent feedback on use. The case was selected for this opportunity to examine decisions in context, rather than as a representative sample of artisan enterprises. Using an exploratory qualitative single-case design, the study contributes an account of how practitioner knowledge informed task selection, information preparation and output evaluation, alongside perceived benefits and further needs after use.

Human judgment and AI capabilities can complement one another in organizational decision making [1], while automation and augmentation remain interdependent [2]. Human–AI complementarity concerns the potential for differences in information and capabilities to support joint work. Hemmer et al. [3] distinguish this potential from its realized effect and from complementary team performance, where joint performance exceeds that of either humans or AI alone. This study uses complementarity as an interpretive lens for examining the contributions of AI and practitioner judgment. It does not test comparative performance. Co-creation concerns participation in creative work, with interaction between human and AI partners central to Rezwana and Maher’s COFI framework [8]. The contribution of AI can also vary with design experience: novices benefited particularly in idea generation, while experienced designers benefited in refinement and output quality in the study by Wang et al. [9]. These findings inform the owners’ concern for creativity but the object of collaboration must be specified. At Gung Mog Kaam, AI-assisted marketing imagery presents an existing ceramic product; it does not, by itself, establish joint design of the ceramic object. Human–AI complementarity therefore provides the broader lens for this study, with co-creation informing the interpretation of its creative activities.

An entrepreneur’s initial assessment of AI involves both domain experience and technological familiarity. These factors appear in Gupta and Yang’s adoption model, which covers pre-perception and perception, assessment and outcome [4] and was subsequently examined using responses from 482 entrepreneurs [10]. Related research addresses AI acceptance intention through 476 responses [11] and AI adoption constructs and digital entrepreneurial intentions among 392 young adults [12]. Together, these studies help frame the assessment that precedes use. Their emphasis on intentions and perceptions, however, leaves a further question for this case: how does the entrepreneur assess specific outputs once AI becomes part of business activities? The conditions of a small business also shape its options for using AI. A review of SME adoption identifies technological, organizational and environmental influences [5], while research on applications across business functions highlights resources, training and implementation support [13]. Adoption and readiness also vary across applications, as examined in a survey of 47 Irish businesses focused on new product development [14]. This literature supports attention to particular activities within the enterprise. Beginning with marketing or customer information while retaining existing approaches to product creation is a task-level choice that a general account of adoption may not fully describe. The outcomes examined in this literature need to be distinguished. At the firm level, AI investment has been associated with revenue growth in a U-shaped relationship in European firms [15]. In handicraft enterprises, a study of e-commerce adoption that included knowledge management found a nonsignificant direct path from AI to adoption [16]. A separate predictive study compared machine-learning models for estimating AI service focus from business attributes [17]. These financial, adoption and predictive outcomes address different questions and cannot be treated as equivalent evidence of benefit. The present case examines how AI was used and evaluated within one enterprise, including benefits reported by its owner after implementation.

Adoption does not resolve how much reliance to place on an output. Prior sector knowledge is relevant to this question: an experiment with 124 entrepreneurs examined AI access alongside such knowledge and reported differences in decision depth, novelty and contextual understanding [6]. In expert work, accounting practitioners also draw on situation-specific knowledge when assessing intelligent automation [7]. The need to evaluate individual tasks is reinforced by evidence that AI assistance can improve performance within its capability frontier but reduce correctness outside it [18]. These studies support examining how business and craft experience inform output evaluation, without assuming that an expert’s judgment is always correct. Expertise can also supply knowledge for an AI application. Research on intellectual capital connects AI with knowledge management and entrepreneurial decision making [19], providing a basis for examining the owner’s interest in a knowledge base. Business descriptions and stories can be organized for customer communication, while the craftsperson continues to interpret their meaning. Articulating this information does not establish that embodied making skills have also been transferred. The case therefore examines two roles for the owners’ knowledge: preparing information for an application and assessing its outputs. Their decision to retain ceramic creation within their own work is considered alongside these uses.

An exploratory qualitative single-case design was used to examine AI use within its business context [20]. The unit of analysis is the enterprise’s selection and evaluation of AI roles during the initial marketing engagement. Its scope covers the documented implementation and subsequent written feedback. Proposed future activities are identified separately. Råbu and Binder’s discussion of qualitative single-case research also informed attention to contextual development and researcher reflexivity [21], although their work concerns psychotherapy. Case selection was purposive: the engagement provided access to craft practitioners introducing AI, their implementation decisions and subsequent reflections. Gung Mog Kaam’s ceramics connect local nature and cultural stories with the owners’ creative work. During consulting conversations documented in field notes, one owner-artisan reported holding a bachelor’s degree in computer science. The university–enterprise collaboration provided a setting for examining how technological familiarity and craft experience informed implementation decisions.

Evidence consists of field notes documenting informal conversations and observations during a university–enterprise consulting engagement, implementation records, marketing outputs and written post-implementation feedback. During these conversations, participants discussed business problems, priorities, task allocation and assessments of AI outputs. They were documented as field notes rather than verbatim research interview transcripts. Implementation records show activities and outputs, while the feedback describes perceived benefits and future plans. Accounts from the conversations are therefore paraphrased. As course instructor and implementation adviser, the researcher guided tool selection, analysis and adaptation to the business. The entrepreneur supplied business information, explained problems and requirements and assessed the applicability of proposed actions. Students contributed to data analysis, search-oriented content planning and chatbot development concepts. This collaboration forms part of the context in which the enterprise adopted AI. The entrepreneur provided verbal and written consent for the use of information documented during the consulting engagement in this study and its publication, including permission to identify the enterprise by name.

Qualitative content analysis began by arranging the evidence into initial motivations, implementation activities, output evaluation and post-use reflection. Passages and documented actions relevant to the research question were then coded descriptively. Codes included incorrect business information, checking contextual fit, retaining product authorship, curating business knowledge and connecting insight to action. Related codes were grouped into the four themes presented below. These themes were interpreted in relation to human–AI complementarity and compared with the other studies reviewed above. During analysis, completed activities were distinguished from proposals and reported benefits from measured outcomes. Evidence that qualified the interpretation was retained, including the entrepreneur’s openness to information beyond personal experience and the capacity constraints affecting workshops. Accounts and outputs were compared for consistency. Because they arose from the same engagement, this comparison offered limited independent corroboration. AI-assisted review analysis was one of the business activities examined, with attention to how it was selected, evaluated and used. The 436 reviews reported below came from several ceramics businesses and served as marketing inputs. They were not a sample of Gung Mog Kaam’s customers.

The owners regarded Gung Mog Kaam as a craft business and placed importance on their own creative contribution. Their initial interest in AI concerned marketing, storytelling and the organization of business knowledge. Ceramic design and pattern creation remained with the artisans. This allocation reflected the owners’ priorities for introducing AI into the enterprise. One owner-artisan’s computer science background and understanding of AI formed part of this context. Familiarity with the technology coexisted with a preference to retain ceramic creation within the artisans’ work. The owner saw opportunities for AI to support the business without immediately applying it to the design of ceramic objects. Marketing imagery illustrates this choice. AI-generated images presented settings in which existing ceramic products could be used, with the actual product retained as the focal object. Drawing on customer and industry insights, the entrepreneur wrote the accompanying text. AI supported the presentation of the work, while the entrepreneur retained responsibility for its written explanation. Automated rewriting, translation, video production and scheduling appeared in development plans, rather than as documented completed activities.

The entrepreneur reported encountering an incorrect AI-generated explanation of the enterprise’s name. Knowledge of the business provided a basis for recognizing the discrepancy. This experience raised a practical concern about how customers might understand the enterprise when obtaining information through AI. In the entrepreneur’s view, AI was becoming a common means of finding business information, including information about craft enterprises. Accurate representation therefore became an immediate priority. The response involved developing a chatbot and preparing a knowledge base from the enterprise’s own information, with LINE Official Account identified as the customer-facing channel. Preparing this information also addressed the owner’s interest in extracting and organizing knowledge held within the business. Descriptions, stories and answers to customer questions could be made available through the application, while their meaning remained grounded in the owners’ knowledge. AI was thus used to help communicate enterprise knowledge, rather than replace all contributions of the craftsperson. This activity concerned information that could be expressed and documented. It does not establish that the embodied skills of ceramic making were transferred to the system. Nor does the case establish whether the chatbot and knowledge base changed descriptions produced by external AI search services.

AI-supported market analysis brought information from outside the enterprise into the marketing process. More than 1,000 Google reviews of several ceramics businesses in northern Thailand were initially collected. Removing star-only ratings and irrelevant entries left 436 reviews. ChatGPT was used to summarize customer jobs, pains and gains. Prompts specified the role, context, task, constraints and output format, including an instruction to distinguish explicitly stated information from inference. ChatGPT’s summaries identified interests in meaningful tourism experiences, making ceramic objects, spending time with others and taking home a reminder of the experience. Concerns included booking information, uncertainty about whether businesses offered workshops or only sold products, waiting times for firing and delivery and value for money. These summaries provided material for considering customer needs and marketing opportunities. NotebookLM was also used to organize seven industry articles. Human synthesis identified storytelling, local identity, handmade value, creative-economy support and niche tourism as relevant themes. The entrepreneur assessed these interpretations against knowledge of the business, its cultural context and what could be delivered in practice. Some outputs did not match the entrepreneur’s experience. Rather than accepting them directly, the entrepreneur questioned their applicability. These disagreements concerned contextual fit and were not necessarily factual errors of the kind reported in the explanation of the enterprise’s name. Practitioner experience served as a basis for assessment, although the case does not independently establish that every judgment made by the entrepreneur was correct. Operational constraints were also relevant. Review-derived insights suggested an opportunity for a beginner-friendly ceramic experience, while the case records noted a temporary workshop closure associated with capacity. A potential market opportunity therefore had to be considered alongside the enterprise’s ability to provide it. This illustrates why identifying customer interest was only part of the decision.

Written feedback after implementation described a clearer connection between customer information, AI analysis and marketing action. The entrepreneur reported that search and social information offered perspectives beyond personal experience. Experience was therefore used to assess new information, while the owner also recognized value in learning from information outside the business. According to the feedback, chatbot made basic enterprise information more accessible. The collaboration also provided learning opportunities: the enterprise gained exposure to tools and perspectives, while students worked with the requirements and constraints of an operating business. These accounts describe perceived benefits rather than independently measured improvements in business performance. For further development, the entrepreneur proposed more systematic tracking of search behavior, engagement, customer questions and feedback. This proposal reflected a need to examine how the applications performed over time. Such monitoring had not yet produced a further documented cycle of evaluation and revision. Table 1 brings together the AI applications described in this section, showing the business needs they addressed and the responsibilities retained by practitioners.

Technological familiarity coexisted with selective AI use in this case. The owners recognized opportunities for AI support while retaining ceramic creation within their own practice. Entrepreneurial adoption research considers technological familiarity, perceptions and acceptance intentions [4, 10–12]. This case adds a contextual account of how willingness to use AI was expressed through decisions about particular activities. Such attention to individual applications is consistent with evidence that adoption and readiness vary across tasks [14]. Retaining creative work therefore should not automatically be interpreted as a lack of technological understanding or rejection of AI. Practitioner knowledge served both as input to an application and as a basis for evaluating its outputs. The owners supplied enterprise information for the chatbot and assessed market insights against business experience and operational capacity. These contributions can be interpreted through human–AI complementarity, particularly the potential arising from differences in information and capabilities [1, 3]. However, the case does not demonstrate superior joint performance. Its contribution is to show how practitioner knowledge entered implementation decisions. Research on sector knowledge and situated expertise supports this interpretation [6, 7], while evidence of task-dependent AI performance reinforces the importance of evaluating specific outputs [18]. The distinction between factual accuracy and contextual suitability is especially relevant: an incorrect business description and an opportunity that exceeds current capacity require different judgments. For culture technology, the case distinguishes AI support for communicating craft from participation in creating the craft object. Co-creation research examines interaction in creative processes [8] and differences associated with design experience [9]. Here, the documented creative collaboration concerned marketing imagery around existing products, while ceramic design remained with the artisans. Preparing descriptions and stories for a chatbot also connects with AI-supported knowledge management [19] but does not establish the transfer of embodied making skills. These observations suggest that support for artisan enterprises should address what practitioners wish to delegate, what knowledge they need to supply and how they will assess the resulting outputs. The findings are limited to one enterprise within a university-supported consulting engagement. Field notes and written feedback provide less detail than complete interaction records and the researcher’s advisory role may have influenced implementation and reported benefits. Chatbot accuracy and commercial outcomes were not independently evaluated. Future research could compare enterprises with different levels of craft experience and technological familiarity and follow changes in task selection over time. Comparing specific AI outputs with practitioners’ assessments would also help distinguish factual errors, contextual disagreements and creative preferences. This study concludes that craft expertise and business experience informed selective AI use through task selection, knowledge preparation and output evaluation. At Gung Mog Kaam, understanding AI did not entail delegating ceramic creation and questioning its outputs did not preclude learning from them. The case makes visible how an artisan enterprise incorporated AI assistance while retaining responsibility for the meaning and creation of its work.