BuyMate: Making AI Interventions Effective in Promoting Rational Consumption in Live Commerce

Paper Title

BuyMate: Making AI Interventions Effective in Promoting Rational Consumption in Live Commerce

Publication Info

• Topic area: AI-driven interventions for rational decision-making in live commerce.
• Keywords: live commerce, impulsive buying, rational consumption, AI intervention, user autonomy, multimodal interaction, sustainable consumption, persuasive technology, responsible AI, decision support.

Background and Problem

• Problem / challenge: Live commerce platforms exploit emotional and cognitive vulnerabilities through manipulative sales tactics, leading to impulsive purchases. Existing AI systems focus on personalized recommendations rather than supporting rational decision-making, and traditional nudges lack adaptability for high-pressure, multimodal environments.
• Significance: Impulsive purchases contribute to financial stress, resource waste, and unsustainable consumption patterns, necessitating tools to empower users to make informed decisions.
• Motivation and related work: Prior research identifies manipulative mechanisms in live commerce but lacks effective counter-persuasion systems. AI-augmented decision-making has shown promise in other domains but remains underexplored in real-time, high-arousal live commerce scenarios.

Solution

• Proposed approach: BuyMate, an AI-driven rational consumption support system, provides real-time, multimodal interventions to counter impulsive buying in live commerce.
• Novelty:
  1. Introduction of a user-centered, “empowerment rather than replacement” approach.
  2. Development of two core modules: Similar Product Comparison and Sales Pitch Recognition and Reframing.
  3. Integration of multimodal, low-interference interventions tailored to live commerce contexts.
  4. Practical insights into designing responsible AI for e-commerce.
• Procedure and key techniques:
  • Conducted formative studies (surveys, interviews, co-design workshops) to identify user needs.
  • Designed BuyMate with two modules:
    • Similar Product Comparison: Presents concise, transparent information on alternative products.
    • Sales Pitch Recognition and Reframing: Detects and reframes manipulative sales tactics in real time.
  • Implemented the system using a large language model (DeepSeek-V3.1) for real-time analysis and intervention.
  • Evaluated through mixed-methods experiments with 35 participants across four intervention conditions (no intervention, human intervention, text-based AI, voice-based AI).

Results

• Concrete findings:
  • AI interventions significantly reduced impulsive purchases compared to no intervention (F = 9.385, p < 0.001).
  • Text-based interventions showed the most stable decision effects, while voice-based interventions were more noticeable and timely.
  • Over 90% of users found the Similar Product Comparison module effective; over 50% rated the Sales Pitch Recognition and Reframing module as effective.
  • System usability (SUS mean = 79.71) and user experience (UEQ dimensions exceeding excellent thresholds) were rated highly.
• Advantage over baselines:
  • AI interventions outperformed human intervention in reducing purchases and fostering rational decision-making.
  • BuyMate’s multimodal approach provided greater autonomy and reduced cognitive burden compared to traditional nudges.
• Experiments / evaluation:
  • Simulated live-stream shopping scenarios with 35 participants across four conditions.
  • Measured purchase behavior, system usability (SUS), user experience (UEQ), and qualitative feedback.
• Limitations and future work:
  • Limited sample diversity (mostly young, educated participants from urban areas).
  • Incomplete detection of implicit sales tactics and potential overreliance on AI.
  • Short-term evaluation; long-term effects on behavior and adaptation remain untested.
  • Future work includes expanding sample diversity, improving implicit cue detection, and conducting longitudinal studies.

Summary

BuyMate is an AI-driven system designed to support rational consumption in live commerce by addressing impulsive buying behaviors. It employs two core modules—Similar Product Comparison and Sales Pitch Recognition and Reframing—to provide real-time, multimodal, low-interference interventions. User evaluations demonstrated significant reductions in impulsive purchases, high usability, and positive user experience. While limitations include sample representativeness and short-term evaluation, BuyMate offers a foundation for responsible AI in e-commerce, promoting sustainable consumption and user autonomy. Future research will focus on enhancing system functionality, expanding user diversity, and validating long-term impacts.