A novel human-AI collaboration framework for service failure recovery and service quality enhancement with LLM-TRIZ integration

Traditional service complaint handling and service quality enhancement rely on human expertise. This research develops a ‘CARE’ model, a staged Human-AI collaboration (HAIC) framework. The four-phase process comprises: (1) complaint coding and cause analysis, (2) AI-Expert dual-path TRIZ reasoning, (3) recovery strategy integration, and (4) evidence-based evaluation. In this model, a Rule-constrained T-CoT prompt embeds a Service TRIZ parameter set, a Service TRIZ contradiction matrix, transformation rules, and a principle-strategy mapping library into LLM reasoning, while Expert-TRIZ and QFD provide contextual validation and requirement-based integration. Applied to 118 taxation service complaints, the model identified 13 service failure causes and generated an integrated set of 20 recovery strategies. For the T-CoT prompt, multi-run testing achieved an F1-score of 0.78 for CR/SF identification and exact and partial contradiction-match rates of 0.62 and 0.84. Under expert-evaluated criteria, the HAIC configuration yielded a 10% redundancy rate, a 95% relevance rate, and a quality score of 4.0 for the generated strategies, reflecting expert judgments of strategy plausibility and conceptual service enhancement. The ‘CARE’ model constitutes a traceable complaint-to-design mechanism for combining LLM exploration with expert oversight in complaint-driven service failure recovery.