As the luxury second-hand market expands,
ensuring accurate sentiment analysis has become a key factor
in maintaining brand trust — one that depends heavily on operational efficiency.
Gugus partnered with Letsur to develop an AI-based emotion recognition system
powered by image analysis.
Gugus’ No. 1 second-hand luxury goods trading platform is an organization that performs tens of thousands of appraisals per year. Gugus’ growth is based on the expertise of experienced appraisers and customer trust built around emotional accuracy. However, as the increase in transaction volume, heightened consumer expectations for authentication, and demand for operational efficiency, a technology-based systematic response that can simultaneously improve the quality and speed of appraisal work has become necessary. In the past, Gugu tried to introduce AI-based emotion technology, but the project was interrupted due to poor data structure and lack of problem definition. As a result, this project went beyond simple model development and was designed to completely review problem definition, data structure design, and product design based on user experience, which are prerequisites for introducing AI technology. Letsur defined this project as the strategic task of building a digital appraisal system for Googus, and carried out a comprehensive approach covering improving the system for using sentiment data, structuring the appraisal work, and designing a product based on user experience. In addition to improving work efficiency in the short term, this is in line with Googus' medium- to long-term goals of establishing AI-based sentiment standards and internalizing the organization's data capabilities in the long term.
First, we investigated all the images, sentiment results, brand information, and serial data accumulated within Gugu and redefined them as a structure for AI learning. Existing data was sufficient for appraisal work, but model learning was highly unstructured and limited in terms of consistency and scalability.
Letsur collaborated with an appraiser to reconstruct key inspection points such as serial location, logo pattern, and wear area with structured input values, and designed a data purification template so that new data can be continuously accumulated. Through this, we have laid the foundation for a circular structure that enables continuous learning and upgrading beyond the level of a single project.
Previously, AI was designed under the comprehensive goal of “AI automatically determines the price”, but the role definition was restructured based on feedback from actual users (appraisers).
Letsur has broken down the emotion process into four steps: ① serial recognition → ② emotion point extraction → ③ anomaly detection → ④ result judgment assistance, and designed specific subtasks that AI can perform at each stage. Through this, AI has been redefined as a function that supports the judgment of appraisers rather than final judges, and it was possible to implement UX that can be easily integrated into actual workflows.
By benchmarking overseas luxury platforms operating successful AI-based appraisal systems, we analyzed key functional flows such as automatic serial detection, automatic generation of analysis reports, and reliability notation.
Based on this, we designed an appraiser-based recommendation function, report automation based on quantitative and qualitative evaluations, and a serial-based pre-filtering function to improve sentiment accuracy as differentiating elements unique to Gugus’. Each feature is structured to contribute to actual utility for users and appraisers.
In order to be implemented as a product that appraisers can actually use, we designed a product unit that goes beyond implementing simple functions. The workflow was analyzed through interviews with appraisers, and the screen flow, input/output structure, and exception handling for each stage were clearly defined.
To increase confidence in AI results, we have implemented a sophisticated user interface that includes providing evidence for result values, requesting manual reviews, and supporting reliability indicators.
Through this project, Guggus went beyond simply introducing AI functions and built the foundation for an emotional support system that is naturally integrated into the actual appraiser workflow. In particular, structuring previously unstructured data assets and designing an AI cycle structure that enables continuous learning based on this is one of the biggest achievements.
Actual effects have also been confirmed at the operation site. In the early stages of the appraisal process, AI automatically analyzed and provided serial recognition and abnormal elements, and the repetitive work time of appraisers was reduced by an average of 40%. Furthermore, by reducing the difference in criteria between appraisers, it is now possible to expect a reduction in the rate of re-appraisal requests.
The automated ability to report appraisal results is also making changes in terms of customer response. By providing both the appraiser's judgment and AI's quantitative analysis results, customers are accepting the appraisal results with higher trust, which is contributing to strengthening the expertise and transparency of the Guggus brand.
The most significant change is that through this project, AI was established not as an independent tool, but as a 'collaborative system' that assists appraisers' judgments and evolves based on data. The records revised and supplemented by the appraiser are stored as automatic logs, which are used as data for future model improvements. As a result, this system can evolve into an active learning-based iterative learning structure, and can continuously improve its responsiveness to exceptional cases.
This project was a turning point for not only developing tools, but also establishing an emotional system where humans and AI collaborate within the organization. This will be a strategic asset for Gugu to lead the future standard of AI-based second-hand luxury goods appraisal.
Beyond simply implementing technology, this project was a structural innovation challenge for an organizational unit that redefined overall appraisal work. The following two implications may serve as essential criteria for similar upgrading projects in the future.
AI should be designed not as a means of automating existing processes, but as a tool to structure the appraiser's workflow together with the organization's judgment system. In this project, we have refined the appraisal process and defined subtasks so that AI can contribute to each stage to enhance applicability in the field.
It is not simply a means for AI learning, but a core foundation that links a circular structure for formalizing sentiment standards, securing consistency in customer response, and upgrading models. Designing data in a form that can be refined, stored, and expanded, as in this case, determines the digital competitiveness of the entire organization.
Letsur will continue to cooperate with companies dealing with data and expertise to create projects that go beyond simple implementation of functions to practical problem solving and organizational change. In particular, in fields where qualitative judgment is central, such as emotions, the design of how technology can change the internal standards and culture of an organization is key. Letsur will refine standards of trust through AI, design a structure where those standards are reversed back into data, and help technology become a core competitive advantage for companies.
.svg)
