Preprint reports simpler AI support workflow system with slower responses

Researchers report in a preprint that a portable set of instructions and tools generated customer-support resolution workflows with 99.2% procedural equivalence to human-approved answers and 83.3% evidence sufficiency. Their deployment case study describes an alternative to complex systems that retrieve documents to ground AI responses. The approach combines natural-language instructions, a retrieval API, format-checking scripts and an AI-generated index over the documents. The authors say implementation took 1-2 engineer days and maintenance involved editing text and data rather than retraining models. Processing took approximately 90 seconds per case versus approximately 8-9 seconds for their production assistant. The findings apply to a mature, well-curated document collection, a workflow that tolerates delays and human review of every case.