OpenSey, Introduces 'AI Synthetic Consumer' Conversation Feature... 'Reduces Costs and Enables Precise Target Analysis'

AI research tech specialist OpenSey (CEO Hwang Hee-young) introduced a 'synthetic consumer conversation feature' implemented with AI technology on its consumer intelligence platform 'Dataspace' on the 12th. Attempts to create digital twins of actual consumers and conduct surveys through them began in the U.S. several years ago. An OpenSey representative noted that while there are attempts to use synthetic responses domestically, there is no existing system that generates synthetic consumers based on massive consumer response data and supports conversation as comprehensively. Now, Dataspace users can interview synthetic consumers based on real consumer data to deeply explore marketing targets. This involves creating segments using OpenSey's own trend report or user survey data and generating synthetic consumer personas representative of those segments for direct dialogue. Specifically, synthetic consumers are created by combining algorithms and large language models (LLMs). Machine learning-based predictive models learn from past survey responses of actual panels to create individual preference data, and LLMs generate evaluations and conversations based on this data. Unlike creating synthetic consumers solely with LLMs, they emphasized adding a statistical prediction model layer to control issues like LLM overfitting and the bias of honest respondents. By responding based on data representing actual panel distributions, they can enhance result accuracy. Synthetic consumers conduct responses based on internal data if available, or infer appropriate responses using persona profiles when data is lacking. Their differentiation lies in clearly distinguishing response bases to provide reliable insights, unlike generic AI generating arbitrary answers. Post-conversation, hybrid reports combining quantitative data and interview content are also provided. They emphasized that synthetic consumers sufficiently reflect actual consumer characteristics and traits. Synthetic consumer reliability is judged based on fidelity (how well synthetic data matches original data distribution), accuracy (similarity between results from synthetic and actual panels), and privacy protection (whether synthetic data replicates specific personal information from original data).