Artificial Intelligence-Powered Customer-Service Chatbots in Small and Medium-Sized Enterprises: A Structured Literature Review of Adoption, Performance, Challenges, and Research Gaps

AI-powered chatbots are widely promoted as a low-cost way for small and medium-sized enterprises (SMEs) to answer routine customer enquiries, yet the evidence specific to SMEs is scattered across information systems, operations and consumer research. This article presents a structured literature review of research on AI-powered customer-service chatbots in SMEs. It applies eligibility criteria fixed before assessment, documents the queries used, appraises each study with design-specific criteria and synthesises the findings thematically. Because access to bibliographic databases was not available, searching relied on a general web search engine and citation tracking, so the review does not claim to be exhaustive or to meet PRISMA 2020 at the identification stage. The evidence set comprises seven empirical studies: two adoption or resistance surveys, one operational case study, one usability study, one qualitative interview study, one mixed-method feature-design and prototype study, and one supply-chain performance survey. In a survey of 292 Fijian SMEs, relative advantage, employee capability, financial resources and top-management support were associated with higher adoption intention, perceived complexity and cost with lower intention, and environmental pressures were not significant. In the one quantified implementation, a Slovak micro-enterprise recorded a 45.9 per cent fall in mean response time and a 14.5 per cent rise in satisfaction, but only the satisfaction change was statistically significant. Usability scores for three deployed chatbots in Ecuador ranged from 72 to 93. Recurring barriers were ambiguous queries, loss of empathy, integration with existing systems, continuing maintenance and the need for human escalation. The article concludes that direct, outcome-based evidence for SMEs is thin, short-term and mostly single-site, and that none of the studies reports evaluating a generative chatbot. It proposes a research agenda covering multi-site field studies, a standard outcome set, total cost of ownership and under-represented languages.