Fast and flexible? Evaluating CLIP for image classification in a marketing context
Аннотация
Artificial intelligence (AI) has accelerated methodological innovation in marketing, particularly through advances in the analysis of unstructured image data. While supervised image classification has facilitated image-based analysis, its reliance on labeled datasets, task-specific training, and substantial computational resources can limit its application. This study evaluates Contrastive Language-Image Pre-training (CLIP)-based models as a flexible and time-efficient alternative for marketing image analysis across three studies. Study 1 compares CLIP-based models with supervised image classifiers to evaluate their predictive performance and computational efficiency without task-specific training. Study 2 uses heterogeneous web-sourced shoe images to examine classification performance under image conditions more representative of real-world marketing data. Study 3 extends the analysis to a more semantically complex sports franchise-brand setting and examines whether domain-specific prompts can improve classification performance. Collectively, the findings suggest that CLIP-based models can provide competitive predictive performance with substantial savings in training time, while model selection and prompt modification provide flexibility for adapting image classification to different marketing contexts.