Small and medium enterprises (SMEs) are pivotal to the economic conditions of emerging markets, including India, Africa, Morocco, and Colombia (Omar Bakar et al., 2020). In this context, SMEs are not only ubiquitous, representing over 98% of all businesses, but also play a crucial role in job creation, employing approximately 70% of the workforce across various sectors (OECD, 2023). This significant presence highlights their integral role in the national economy and exhibits trends observed in other emerging markets (Breuer et al., 2023), indicating a common pattern of economic dynamics in such environments (Mishra & Khanna, 2024). However, SMEs in emerging markets contribute just 39% to the gross domestic product (GDP) (Pratikto et al., 2024), markedly less than the 57% of developed markets (European Commission, 2022). This disparity indicates untapped potential and requires enhanced business interventions to boost the economic contribution of SMEs (Amoah et al., 2022).

The conventional approach to SME capacity building in emerging markets has been predominantly focussed on human training interventions (Mendy, 2021). These interventions have been key to providing knowledge and guidance (Ferreira et al., 2021), strengthening growth and development across different business domains (Msimango-Galawe & Majaja, 2022). When personalised, interventions can be especially beneficial in addressing the unique challenges and aspirations of individual enterprises, although this traditional model has some constraints (Idris et al., 2023). Personalised human business training grapples with scalability and accessibility, especially in a heterogeneous sector and the geographically widespread SME landscape (Pfister & Lehmann, 2021). Therefore, costs associated with tailored training programmes hinder growth of many SMEs, limiting their access to vital developmental resources and opportunities (Baloyi & Khanyile, 2022).

To overcome this challenge, the advent of artificial intelligence (AI) coaching emerges as a promising and innovative solution (Alhammadi, 2023). Researchers like Chowdhury et al. (2023) and Terblanche et al. (2024) agree that AI coaching embodies the convergence of technology and personalised training. Artificial intelligence coaching is designed to affordably scale bespoke learning experiences, which are essential in translating complex theoretical knowledge into actionable and effective business strategies (Agatova & Latipova, 2024; Terblanche et al., 2022a). The amalgamation of the intuitive, empathetic qualities of human coaching with the scalability of AI coaching to personalise the training experience presents a potentially ground-breaking approach (Bagai & Mane, 2023). This synthesised model holds immense potential to revolutionise the landscape of personalised SME training affordability, aligning with the dynamic and evolving needs of SMEs in emerging markets (Raihan, 2024).

Human business training and AI coaching are vital for facilitating SME development and sustainability, but research on their combined impact is scarce (Berretta et al., 2023). Furthermore, the existing literature often separates human-centric training from AI-driven coaching, missing their potential synthesis (Perifanis et al., 2023). This failure to explore the integration of human and AI approaches is pronounced in emerging markets where SME operational patterns differ from developed countries (Zahoor et al., 2023).

Therefore, this study sought to understand how an AI coach facilitated the post-training application of business knowledge for SME owners and managers. Specifically, it aimed to identify the functionalities of AI coaching that participants perceived as beneficial for knowledge application, whilst also exploring the methodological and adoption challenges that mediated its effectiveness.

Small and medium enterprises majorly contribute to economic development and job creation, particularly in emerging markets (Mer & Virdi, 2024). Yet many SMEs struggle to convert training-derived knowledge into business improvement. A review by Leger et al. (2025) reported inconsistent outcomes from entrepreneurship training in sub-Saharan Africa, showing how difficult knowledge application can be.

Weak knowledge resources and poor learning networks constrain SME innovation (Durst et al., 2022). Generic training often overlooks sector diversity, limiting relevance (Chikweche & Bressan, 2018). Consequently, learning is mostly experiential and informal, underlining the need for scalable and context-sensitive support mechanisms (Dekel-Dachs et al., 2020). To design such support, it is first necessary to understand the underlying mechanisms of adult learning and knowledge transfer.

Kolb’s (1984) experiential learning model explains knowledge acquisition as a cycle of experience, reflection, conceptualisation, and experimentation. Kolb’s model is particularly applicable to this study because to its focus on experiential and action-oriented learning. This directly relates to the practical application of business training by entrepreneurs (Motta & Galina, 2023). Furthermore, learner motivation is essential for this cycle to succeed, with self-determination theory stating that supporting an individual’s innate needs for autonomy, competence, and relatedness strengthens the intrinsic motivation required to persist in applying new knowledge (Deci & Ryan, 1985). Organisational support improves the likelihood that learning is applied (Salamon et al., 2021). However, without reinforcement, most training is lost in the workplace (Cromwell & Kolb, 2004). Structured post-training support is therefore important (Mansour et al., 2022). One of the most established forms of such structured support is workplace coaching.

Workplace coaching helps employees integrate training into practice (Kapoutzis et al., 2023; Régnier, 2023). Meta-analyses show that coaching improves performance, well-being, and learning transfer (Dixit & Sinha, 2022; Jones et al., 2016; Park et al., 2021). Yet human coaching is expensive and difficult to scale, particularly for SMEs (Passmore et al., 2025a). Artificial intelligence-supported coaching provides a potential solution (Terblanche et al., 2022a).

Artificial intelligence coaching employs chatbots and conversational agents to set goals such as applying a new marketing technique, give feedback by tracking progress against that goal, and support behaviour change by prompting reflection on challenges (Graßmann & Schermuly, 2020). Studies report effectiveness across organisational settings (Passmore et al., 2025a). Terblanche et al. (2022b) found AI chatbot coaching, in a study of new graduates transferring learning to work tasks, produced goal-attainment comparable to human coaching, whilst Terblanche and Tau (2024) showed that an AI goal-attainment chatbot helped new graduates transfer learning to work tasks.

Therefore, AI coaching tools can complement human coaching by managing routine elements and enabling facilitators to focus on complex learning needs (Mai et al., 2021; Passmore & Woodward, 2023). By guiding users through questioning, feedback, and progress-tracking, AI coaching operationalises key elements of experiential learning such as reflective observation and active experimentation (Kolb, 1984). This suggests that AI coaching can provide cost-effective, scalable post-training support linking adult learning theory with practice, particularly where human coaching access is limited. However, its effectiveness also depends on clear communication about the AI’s role and its alignment with learning goals (Engström et al., 2024). Furthermore, whilst much research has focused on AI in formal education or goal-tracking, its specific role in reinforcing the informal, post-training experiential learning cycle for entrepreneurs remains underexplored.

Adoption depends on user perceptions. The Technology Acceptance Model (TAM) (Davis, 1989) links acceptance to usefulness and ease of use, and the Unified Theory of Acceptance and Use of Technology (UTAUT) enhances social influence and facilitating conditions (Venkatesh et al., 2003). The Artificial Intelligence-Technology Acceptance Model (AI-TAM) model introduces trust, output quality, and collaborative intention (Baroni et al., 2022). The perceptions of AI assist in learning and adoption trajectories (Engström et al., 2024). Terblanche and Kidd (2022) applied TAM and UTAUT to a reflective coaching chatbot, finding that age and gender affected intentions, underscoring trust and perceived usefulness as key for coaching tools. Together, these frameworks and studies highlight that beyond functionality, the design of AI coaching systems must prioritise user trust, perceived usefulness, and ease of use to ensure they are adopted and sustained.

This study qualitatively evaluates how AI coaching, combined with human-facilitated training, supports SME leaders in applying business knowledge. Artificial intelligence coaching in organisational practice is still a new area, with recent reviews noting a limited peer-reviewed evidence base (Bachkirova & Kemp, 2024; Passmore et al., 2025a). Research on human–computer interactions demonstrates the use of large language model chatbots for executive coaching, which has only recently been examined in controlled studies, underlining the novelty of this research field (Arakawa & Yakura, 2024). For SMEs, studies on AI adoption report a fragmented body of knowledge and little large-scale evidence, supporting exploratory research at the intersection of AI coaching and SME leadership (Schwaeke et al., 2024). A qualitative method was therefore appropriate because a qualitative method allows the careful study of lived experience, beliefs, and reflection in participants’ own words and supports in developing a rich conceptual understanding in emerging areas (Braun & Clarke, 2019).

An exploratory qualitative design was chosen to analyse the interaction between AI coaching and business training in supporting SME leaders, as this approach allows careful exploration of a new organisational practice in the specific environment in which participants operate (Palinkas et al., 2015).

The study was conducted with SME leaders attending business training shown to correlate with improved organisational performance. Empirical large-scale UK data demonstrate that both on-the-job and off-the-job trainings significantly positively associate with SME performance, as reported by owner-managers (Idris et al., 2020). Similarly, research in Ghana shows that accessible, content-rich managerial training instils effectiveness in SME managers (Atiase et al., 2023). Therefore, the workshop was selected because it provides goal-setting, actionable content, and accessible delivery, which are identified as critical features for meaningful training outcomes.

The workshop presented a structured marketing framework using concise theory, practical case studies, and facilitated peer breakout discussions, enabling participants to test ideas against real challenges and set specific goals for application (Busso et al., 2023). The workshop covered key business skills including digital marketing and customer relationship management. At the close, the goal-attainment AI coaching chatbot was designed as a post-training reflective tool to help achieve personal goals identified during the session (Terblanche, 2019).

After the workshop, participants engaged with the AI coach on WhatsApp. Developed under the Designing Artificial Intelligence Coaches (DAIC) framework (Terblanche, 2020), the AI coach applied goal-setting theory using a generative AI conversation model, maintaining a nondirective flow and structured progress-tracking dialogues.

Dialogue followed a text-based nondirective goal-attainment protocol using open questions, reflection prompts, and twice-weekly progress checks. For example, rather than providing a direct solution, the AI coach would guide the user’s own reflection:

This structured protocol differentiates AI coaching from a generic text helpline by embedding coaching microskills: reflection, goal clarity, and progress-tracking, rather than ad hoc tips.

Following business skill training, an email invitation was sent to all attendees. From those who expressed interest, 20 SME owners or managers (see Table 1) engaging with the AI coach for 4 weeks were selected via purposive sampling. All participants had at least 2 years of business experience and were actively involved in managing or growing a small business in an emerging market. Participant SMEs operated in resource-constrained environments typical of emerging markets, where access to consistent human coaching was limited. All participants were introduced to the AI coach as part of the training programme and were supported in setting it up and encouraged to use it regularly during the engagement period.

The business training occurred in Week 1, followed by a 4-week AI coaching engagement (Weeks 2–5). The semistructured interviews were conducted in Week 6.

Primary data collected by semistructured interviews formed the core of data collection. Each participant engaged in a 55-min (ranging from 45 min to 70 min) online interview conducted using Zoom at the end of the 4-week AI coaching period. All interviews were carried out by the first author. Semistructured interviews are valuable for exploring participants’ perspectives whilst allowing flexibility for follow-up questions (Kallio et al., 2016). Questions included their use of the AI coach, application of training knowledge with the AI coach’s assistance, challenges faced, and their perceptions of the experience’s value. All interviews were audio-recorded with participants’ consent and transcribed verbatim (DiCicco-Bloom & Crabtree, 2006).

The primary data for this study comprised semistructured interview transcripts, which were thematically analysed to provide insights into participants’ self-reported experiences. In addition, data were collected from the AI coaching tool’s chatbot logs. These logs were not thematically analysed and were used solely to validate participants’ engagement with the intervention, such as confirming the number of interactions reported in Table 1. These complementary data sources strengthened the trustworthiness of the findings by corroborating self-reported engagement levels (Fetters et al., 2013).

Data from the semistructured interviews were analysed using an inductive (bottom-up) thematic analysis approach, guided by the six-phase process described by Braun and Clarke (2006). The primary data corpus consisted of the verbatim transcripts from all 20 interviews.

The first phase, familiarisation, involved reading and re-reading the transcripts to gain a holistic sense of the dataset. Following this, a systematic open coding process (Phase 2) was conducted using ATLAS.ti. This produced 202 initial codes capturing participants’ explicit statements, implied expectations, frustrations, and reflections.

The analysis then proceeded through an iterative and recursive process of clustering (Phase 3). These initial 202 codes were collated into 31 conceptual code groups based on shared properties. With researcher discussion and the use of visual mapping, these groups were further refined, merged, and abstracted to generate candidate themes. These candidate themes were critically reviewed and refined (Phase 4) against the coded extracts and the full dataset to ensure they were coherent, distinct, and accurately represented the patterns in the data.

This refinement process resulted in eight coherent subthemes. In Phase 5, this hierarchy was finalised by organising the eight subthemes within the three overarching main themes that form the structure of the findings. The final phase (Phase 6) involved interpreting these themes in the context of the study’s purpose, moving beyond summary to examine what the experiences revealed about the potential and limitations of AI coaching in this SME training environment. The analysis was performed inductively, without a pre-existing theoretical framework.

To ensure trustworthiness (Lincoln & Guba, 1983; Nowell et al., 2017), credibility was fortified using two practical measures. Firstly, transcript validation was performed by the interviewer, who listened back to the audio recordings whilst reading the transcripts to ensure accuracy. Secondly, communicative validation was carried out during the interviews, whereby the interviewer would periodically summarise a participant’s point and ask, ‘Have I understood that correctly?’ to confirm the interpretation in real-time.

Dependability was addressed by maintaining a detailed audit trail within Atlas.ti, which consists of all analysis files, including the final code lists, analytic memos, and thematic maps illustrating the refinement process. Transferability is supported by the thick description of the participant cohort, the emerging market context, and the AI coaching intervention.

Finally, confirmability was addressed through reflexivity. The first author, who conducted the interviews, maintained a reflexive journal to note methodological decisions and bracket potential biases. As part of this, we disclose that the first author was involved in delivering business training and the second author in the chatbot’s technical development. This potential bias was reduced by adhering to the inductive analysis protocol and using researcher discussion to ensure themes were demonstrably grounded in the participants’ data rather than the researchers’ pre-existing knowledge.

Ethical clearance was granted by the Stellenbosch University Institutional Review Board (Ref: SBER 30881). All participants provided written informed consent prior to the interviews, were assured of confidentiality through anonymisation, and were aware they could withdraw at any time. Participants were not previously known to the interviewer.

The first theme examined how AI coaching provides real-time support, fosters reflective thinking, and creates a safe learning environment. This main theme emerged from subthemes such as scalability and real-time responsiveness, fostering reflective thinking, and trust through nonjudgemental support.

This second theme focussed on how intuitive onboarding, transparent data practices, and iterative improvements impact AI coaching adoption. This theme consists of the subthemes: effective onboarding for AI coaching, trust-building through transparent data practices, and consistent evaluation and iterative improvements.

Theme 3 explored platform familiarity, cultural relevance, and engagement strategies that influence AI coaching effectiveness and consisted of the following subthemes: platform familiarity and accessibility and cultural sensitivity and contextual alignment.

Artificial intelligence coaching replicates the benefits of human coaching, such as promoting goal-directed behaviour and reflective feedback, making it a transformative tool for SME development. Jones et al. (2016) demonstrated that coaching enhances performance and wellbeing, whilst Dixit and Sinha (2022) highlighted its role in behavioural skill transfer. In this study, the AI’s nondirective questioning and psychological safety enabled SME owners to apply training in practice by creating a space where they felt comfortable discussing mistakes. Graßmann and Schermuly (2020) support this, showing AI can replicate structured dialogue that promotes behaviour change, which is valuable where resource constraints limit personalised coaching (Pfister & Lehmann, 2021). Terblanche et al.’s (2022b) finding that AI coaching matches human coaching in goal-attainment is mirrored in participants’ ability to explore ideas after hours. This scalability addresses cost and scheduling barriers (Baloyi & Khanyile, 2022), positioning AI coaching as a democratising force for SME training. Yet repetitive feedback echoes Terblanche et al.’s (2024) warning that AI must evolve with users to sustain value. Future designs should include memory-informed dialogues to enhance entrepreneurial impact (Passmore & Woodward, 2023). Artificial intelligence coaching can thus extend the benefits of traditional coaching to SMEs, providing consistent, accessible reinforcement that lengthens the effect of training interventions.

Trust and contextual relevance are pivotal for AI coaching’s effectiveness in SMEs. Participants’ hesitation to share sensitive business information without clear data assurances aligns with Eden et al.’s (2024) observations on transparency’s role in AI adoption. Transparent data practices are essential for SMEs, where strategic information is sensitive (Raman et al., 2025). Additionally, the AI’s language and tone, often perceived as overly formal or foreign, reduced reflective depth, supporting Berretta et al.’s (2023) warning about design biases in AI systems. Multilingual and culturally resonant AI coaching, as proposed by Dara and Kesavan (2024), would enhance engagement, particularly in diverse emerging markets like South Africa.

Kolb’s (1984) emphasis on meaningful reflection highlights the need for contextually relevant feedback. Cromwell and Kolb (2004) argue that continuous, resonant reinforcement drives behavioural change, a process hindered when AI feedback feels misaligned. For SMEs, where experiential learning is often informal (Chikweche & Bressan, 2018), culturally attuned AI coaching could bridge the gap between training and application, fostering entrepreneurial innovation. This study shows the importance of identifying these adoption barriers and being sensitive to culturally responsive AI designs.

The use of WhatsApp as a delivery platform improved AI coaching adoption in this study by using a familiar, low-data tool. This supports Rozmi et al.’s (2021) finding that SME learning grows when delivered through accessible platforms. Terblanche and Kidd (2022) show that facilitating conditions predict technology adoption, noting that the platform is one of three main adoption factors. WhatsApp’s place in SME routines reflects this point. The approach also answers Idris et al.’s (2023) concerns about scalability, allowing SMEs in low-resource settings to take part without technical barriers. Participants valued the short learning curve and low data costs, which made engagement easy.

However, platform limitations, such as difficulty tracking past interactions, hindered long-term engagement, reflecting Terblanche’s (2020) noted trade-off between simplicity and functionality. Future AI coaching designs should balance accessibility with progress-tracking features to support SME owners’ need for structured learning, as Suravi (2023) advocates for hybrid training models. This study shows that familiar technology can democratise access to learning for SMEs, offering a practical solution to scalability challenges in emerging markets.

Participants emphasised the importance of AI coaching that adapts over time, including follow-ups on progress, highlighting the need for systems that evolve with SME learners. Salamon et al. (2021) show that adaptive support sustains learning transfer, whilst Cromwell and Kolb (2004) stress that reinforcement must build on prior experience. Incorporating memory and follow-up features, as recommended by Passmore et al. (2025b), would align AI coaching with human coaching’s longitudinal awareness and strengthen its value for SME growth.

The three themes align with AI-TAM constructs: Functionality relates to perceived usefulness and output quality; methodology involves ease of use, facilitating conditions, and trust; and adoption concerns social influence, behavioural intention, and contextual moderators (Baroni et al., 2022; Venkatesh et al., 2003).

Finally, the findings of this study show that AI coaching enables SME leaders to apply training in real-world contexts by offering immediate, on-demand support, encouraging deeper reflection, and creating a nonjudgemental environment for learning. This aligns closely with Kolb’s (1984) experiential learning cycle, which describes learning as a process involving concrete experience, reflective observation, abstract conceptualisation, and active experimentation. Figure 1 illustrates how AI coaching integrates with this cycle, enhancing its theoretical and practical implications for SME development.

Concrete experience: The workshop gave SME practice grounded in real challenges. Artificial intelligence coaching added immediate guidance, aiding agile decisions (Msimango-Galawe & Majaja, 2022).

Reflective observation: Artificial intelligence coaching encouraged critical thinking and honest reflection, crucial where informal learning prevails (Chikweche & Bressan, 2018; Park et al., 2021).

Abstract conceptualisation: Adaptive feedback supported reframing strategies, even in weak knowledge networks (Dekel-Dachs et al., 2020).

Active experimentation: WhatsApp delivery allowed testing ideas in daily routines. Culturally aligned prompts, as Zahoor et al. (2023) advise, could enhance relevance.

In conclusion, AI coaching creates a continuous learning loop tailored to SMEs, addressing their distinct constraints, such as limited access to human coaching and diverse operational needs (Idris et al., 2023). It extends adult learning theory by showing how technology can support experiential learning in resource-constrained contexts, offering a novel framework for SME training interventions.

The findings also resonate with Knowles’ (1978) andragogy principles, which emphasise self-directedness and problem orientation in adult learners. Artificial intelligence coaching’s user-paced, on-demand support empowered SME owners to tackle context-specific challenges, reinforcing Mansour et al.’s (2022) findings that learner control enhances training transfer. Unlike traditional training, where Cromwell and Kolb (2004) noted limited post-training support, AI coaching’s real-time availability overcomes accessibility barriers, making it a scalable solution for SMEs in emerging markets.

This study contributes theoretically, empirically, and practically to adult learning, coaching science, and entrepreneurial development in emerging market SMEs. The primary novelty lies in demonstrating how AI coaching, when paired with structured human training, facilitates the application of business knowledge amongst SME entrepreneurs. Unlike many previous studies that examined AI tools in isolation, this research integrates AI coaching with experiential learning theory and reveals how learners use digital tools to extend the impact of traditional business training interventions (Chen et al., 2024).

Theoretically, this study contributes to the adult learning discourse by showing that AI-enabled environments could reinforce reflective observation and abstract conceptualisation, two key stages in the experiential learning cycle (Kolb, 1984). Whilst principles of andragogy emphasise self-directed learning (Knowles, 1978), this research adds empirical weight to the idea that AI coaching can enhance learner autonomy through nonjudgemental, user-paced support. Additionally, it supports the notion that control over the learning environment strengthens confidence and motivation (Deci & Ryan, 1985). The findings also affirm that learning systems must evolve with the learner to remain effective (Salamon et al., 2021).

Practically, this study offers direct insights for a number of stakeholders. Training providers and development agencies can integrate AI coaching tools to extend the lifecycle of their interventions and enhance knowledge transfer. Artificial intelligence developers and ed-tech companies can use the study’s findings to build culturally attuned, memory-informed, and user-responsive AI coaching systems. Policy makers may leverage AI coaching as a cost-effective method for scaling SME support across rural or under-resourced areas. Small and medium enterprise support institutions such as incubators and nongovernment organisations (NGOs) can offer AI coaching to supplement mentoring programmes.

Furthermore, for SME owners in emerging markets, this study offers an accessible, self-paced, and psychologically safe support mechanism that helps sustain training outcomes. By enabling immediate feedback, reflective questioning, and goal-tracking, AI coaching assists entrepreneurs in contextualising their learning and addressing daily operational challenges. The capacity to receive ongoing support without cost or scheduling constraints positions AI coaching as a democratising force in SME development.

This study has several limitations that warrant consideration. Firstly, the sample was limited to 20 participants from a single South African cohort, which may affect the transferability of findings to other emerging market contexts with different infrastructural or linguistic conditions.

Secondly, the 4-week engagement period was sufficient to observe initial application but restricted conclusions about long-term behavioural change or sustained business impact. Longitudinal studies are needed to evaluate sustained outcomes.

Thirdly, whilst the use of WhatsApp as a delivery platform enhanced accessibility, its technological constraints, such as difficulty retrieving past interactions, might have inhibited a deeper learning cycle and masked the coach’s full potential.

Fourthly, the study did not collect baseline data on participants’ prior familiarity or comfort with chatbot technology. This is a potential confounding variable, as pre-existing digital literacy could have influenced perceptions of the tool’s utility and ease of use.

Finally, the AI coach was text-based and supported English only. Participants noted this limited cultural resonance, and future research should explore multilingual platforms to enhance inclusivity and reflective depth. These limitations provide boundaries for the interpretation of the findings and highlight clear directions for future research.

Building on this study, several avenues for future research are proposed. Longitudinal studies are needed to evaluate the sustained impact of AI coaching on SME performance metrics, such as revenue growth, customer retention, or team development. Such studies could determine whether short-term knowledge application observed here leads to long-term behavioural transformation.

Additionally, future research should explore AI coaching in multilingual and multicultural contexts. The findings suggest that language and cultural alignment could affect engagement with the AI coach. Investigating how AI coaching tools can incorporate local dialects and culturally resonant phrasing would help address barriers to reflective depth and adoption in diverse settings.

Furthermore, the study revealed unique insights into the role of psychological safety in AI coaching. Participants frequently reported that the nonjudgemental nature of the AI coach enabled them to be more honest about their challenges and business decisions. This created a safe space for vulnerable self-reflection, an outcome rarely explored in technology-mediated learning. Future studies could investigate how design features such as tone, anonymity, and perceived neutrality contribute to psychological safety and how these elements affect learning transfer and user confidence.