When hospitality students learn how to analyse customer feedback, they often face a major challenge: there simply isn’t enough diverse, real-world data to practise with. Gathering genuine customer responses takes time, money, and cooperation from businesses that may be hesitant to share their data.
A new study from the University of Chester aims to change that reality. In a recent paper published in Technology, Knowledge and Learning, Prof. Kelvin Leong and Dr. Anna Sung introduce an artificial intelligence (AI) system called AIRSim (AI Responses Simulator), designed to generate realistic and diverse simulated customer feedback for educational purposes.
Their research article, titled Introducing AIRSim: An Innovative AI-Driven Feedback Generation Tool for Supporting Student Learning, explores how AI can help hospitality students improve their analytical and decision-making skills through simulated data. It also reflects a growing movement within education that integrates AI-driven learning tools to bridge the gap between theory and practice.
Turning customer feedback into a learning tool
Customer feedback has long been a central component of business success in the café and restaurant industry. Surveys, comment cards, and digital reviews all provide insight into customer satisfaction, service quality, and dining experiences. However, as Prof. Leong notes, educators struggle to provide students with sufficient varied data to analyse effectively.
Without access to large, authentic feedback datasets, students often learn only basic principles of analysis rather than gaining experience in handling complex, real-world scenarios. “In hospitality education, diversity in feedback is vital,” the authors write. “Students must understand the range of customer perspectives to make meaningful improvements in service and operations.”
AIRSim addresses this problem by utilizing Generative AI to generate hypothetical feedback datasets that closely mirror real customer interactions. The system generates detailed responses to user-uploaded questionnaires, providing educators with a wealth of material to train students in survey interpretation and data analysis.
How AIRSim works
AIRSim is built on OpenAI’s GPT-based large language model, leveraging the advanced natural language processing capabilities of systems such as ChatGPT. The researchers designed AIRSim to simulate customer responses based on uploaded questionnaires and specified demographic profiles, including nationality, age and gender.
Once an educator uploads a questionnaire, the tool generates a user-specified number of simulated participants. Each simulated participant provides unique and contextually appropriate answers, which are exported to an Excel dataset for further analysis. This structure allows educators and students to perform statistical analysis, identify trends, and practise feedback interpretation within a controlled digital environment.
Unlike direct use of ChatGPT, AIRSim provides a structured interface that automatically produces data suitable for analysis. It eliminates the need for programming or external APIs, making it accessible to educators without technical expertise. According to the study, this practical design makes AIRSim particularly valuable for fields where realistic feedback data is difficult to obtain.
Experimenting with simulated data
To assess the reliability and diversity of AIRSim’s generated feedback, the researchers conducted 16 experiments. These experiments tested the tool’s performance across different questionnaire lengths and participant sample sizes. Each test measured how well AIRSim simulated diverse responses and whether its output resembled the variability of real-world feedback.
The evaluation relied on entropy analysis, a method derived from information theory that quantifies diversity within a dataset. Higher entropy values represent greater variability, while lower values indicate uniformity. In the study, AIRSim consistently produced high entropy scores, meaning that the simulated feedback demonstrated substantial variation across demographic and question-based factors.
Interestingly, the experiments revealed an inverse relationship between the number of participants and the entropy index. Smaller datasets (such as those with ten simulated respondents) tended to show higher diversity, while larger ones appeared more uniform. This finding suggests that, just as in real surveys, sample size influences the distribution of opinions and responses.
AI-driven learning and adaptability
The implications of AIRSim extend far beyond the hospitality sector. As Prof. Leong emphasises, the tool demonstrates how Generative AI can enhance education through data simulation and scenario-based learning. Students using AIRSim engage in experiential learning, analysing data that mimics authentic feedback rather than hypothetical textbook examples.
By examining simulated datasets, students can explore statistical diversity, sentiment trends, and customer behaviour patterns. This approach cultivates analytical thinking, problem-solving, and adaptability skills increasingly valued in service industries. Moreover, exposure to diverse feedback trains students to respond effectively to a range of customer expectations and cultural perspectives.
Such training is especially relevant in the digital transformation of education, where AI-powered tools are helping institutions create scalable, personalised learning environments. By integrating AIRSim into their teaching, educators can simulate hundreds of unique customer experiences, allowing learners to practise interpreting open-ended feedback at scale.
Ensuring reliability and quality
For AIRSim to function as an effective educational tool, its simulated feedback must resemble real-world data. To verify this, the authors employed a combination of pilot testing, human validation and statistical entropy analysis.
During pilot tests, AIRSim generated simulated responses across different scenarios and demographic profiles. Human evaluators then compared these outputs with authentic customer feedback to ensure realism and contextual appropriateness. The diversity confirmed by entropy analysis strengthened confidence in the model’s reliability.
However, the researchers acknowledge that simulated feedback cannot completely replace genuine data. They recommend combining AIRSim-generated responses with real feedback whenever possible to ensure a balanced learning experience.
Bias and ethical considerations
Like all AI-driven systems, AIRSim operates within the limitations of the data that trains large language models. Prof. Leong cautioned that AI-generated content may reflect underlying societal or cultural biases present in those datasets. Such biases can manifest subtly in simulated feedback, influencing how demographic groups are represented.
For instance, responses associated with gender or nationality might inadvertently reflect stereotypes. To mitigate this, educators should review and validate AI-generated data before using it in training exercises. The authors also encourage future developers to refine algorithms to reduce such risks, emphasising that human oversight remains critical for responsible AI use in education.
The future of education lies in blending human experience with AI-powered tools, creating opportunities for students to learn in deeper and more flexible ways.
-Kelvin Leong
From restaurants to hospitals
Although AIRSim was developed for café and restaurant management education, its methodology could be applied to numerous fields where feedback analysis plays a crucial role. In healthcare, for example, simulated patient feedback could help medical trainees learn how to analyse satisfaction surveys or understand patient experiences.
In retail or tourism, AI-generated responses could allow students to practise analysing customer reviews and sentiment data before handling live feedback. Similarly, in finance and education, simulated feedback could support training in client relations or student engagement.
The authors argue that this scalability underscores the transformative potential of AI in professional education. By bridging the gap between limited real data and the need for experiential learning, AI simulators like AIRSim can make training more accessible, efficient and responsive to real-world dynamics.
Future research and development
While AIRSim represents a significant step forward, the researchers highlight several directions for future work. Enhancing the realism of AI-generated feedback remains a priority. Future iterations of AIRSim could incorporate sentiment analysis, natural language emotion detection and adaptive learning algorithms to produce even more nuanced responses.
Prof. Leong also suggests that comparative and longitudinal studies could measure how students trained with simulated feedback perform compared to those who work only with real customer data. Such studies would help determine whether AI-generated feedback leads to lasting improvements in analytical and critical-thinking skills.
Reference
Leong, K., & Sung, A. (2025). Introducing AIRSim: An innovative AI-driven feedback generation tool for supporting student learning. Technology, Knowledge and Learning. https://doi.org/10.1007/s10758-025-09835-9
