{"url":"https://aisel.aisnet.org/cgi/viewcontent.cgi?article=1180&context=icis2025","title":"AI Recovery for Chatbot Failures: Static vs Generative","domain":"aisel.aisnet.org","imageUrl":"https://images.pexels.com/photos/7709141/pexels-photo-7709141.jpeg?auto=compress&cs=tinysrgb&h=650&w=940","pexelsSearchTerm":"chatbot customer service","category":"Other","language":"en","slug":"f2ebe0ec","id":"f2ebe0ec-dba6-4526-a33c-2f71d300c880","description":"Researchers compare static and generative AI recovery strategies for chatbot service breakdowns in customer interactions.","summary":"## TL;DR\n- Researchers compare static and generative AI recovery strategies for chatbot service breakdowns in customer interactions.\n- Repair strategy (action-focused) boosts recovery quality more than Inform (explanation-focused); generative AI aids complex cases but raises user effort.\n- Findings guide chatbot design by favoring adaptive recovery sequences over single static methods.\n\n## The story at a glance\nThis short paper from ICIS 2025 examines recovery tactics after chatbot failures in customer service. Authors Dennis Benner (University of Kassel), Andreas Janson (Institute of Information Systems and Digital Business), and Chee-Wee Tan (Hong Kong Polytechnic University) test static approaches in Study 1 and generative AI in Study 2. It appears now as part of proceedings from the International Conference on Information Systems, held amid rising chatbot use in service.\n\n## Key points\n- Conversational breakdowns harm service quality and brand trust, prompting need for effective recovery.\n- Study 1 (dual-study design) tests two static strategies: **Inform** (explanations only) shows limited gains; **Repair** (action-oriented) significantly improves recovery quality and perceptions.\n- Study 2 deploys generative AI chatbot for interactive, nuanced recovery, yielding higher resolution in complex scenarios.\n- Generative AI demands more user cognitive effort compared to static options.\n- Proposes sequencing static and generative recovery for optimal outcomes under varying conditions.\n\n## Details and context\nThe paper fits the Human Technology Interaction track at ICIS 2025, focusing on real-world chatbot limits like comprehension failures or off-topic drifts.\n\nStudy 1 likely uses experiments to isolate strategy effects, with Repair outperforming due to direct problem-solving.\n\nStudy 2 highlights generative AI's edge in handling ambiguity, but notes trade-offs in user load—relevant as services scale AI for efficiency.\n\nPreliminary results stress condition-specific use, avoiding one-size-fits-all recovery.\n\n## Key quotes\nNone available from accessible metadata.\n\n## Why it matters\nChatbot breakdowns affect millions of daily service interactions, influencing customer retention across industries. Designers and firms gain tools to pick recovery based on complexity, blending static speed with AI adaptability for better satisfaction. Watch empirical validations in live deployments, as preliminary findings may evolve with larger samples.","hashtags":["#ai","#chatbots","#customer","#service","#human-ai","#interaction"],"sources":[{"url":"https://aisel.aisnet.org/cgi/viewcontent.cgi?article=1180&context=icis2025","title":"Original article"}],"viewCount":2,"publishedAt":"2026-04-08T13:02:40.485Z"}