AI TrackerFrontier desk · AEST
Back to news
normalaitech

Cost-Aware Abstention for LLM-Based IT Ticket Classification

arXiv cs.AI · 23 Oct 2026, 03:58 AEST

The study evaluates cost-aware abstention for IT ticket classification using large language models. It finds that a TF-IDF with logistic regression approach significantly reduces costs and wrong routing rates compared to other methods.

Key facts

Read aloud

The study evaluates cost-aware abstention for IT ticket classification using large language models. It finds that a TF-IDF with logistic regression approach significantly reduces costs and wrong routing rates compared to other methods. Wrong automatic routing of an IT support ticket can cost more than human triage. Under the illustrative cost model c ok, c triage, c wrong = zero, one, five, TF-IDF with logistic regression and a threshold of zero point six zero selected on calibration reduces test cost from zero point seven six three to zero point four seven two, a thirty eight point one percent relative reduction. Frozen calibration-selected policies also beat both always-route and always-defer on held-out test data in fourteen of sixteen settings in a post-hoc cost-parameter analysis. The strongest prompted configuration reaches zero point four seven seven accuracy on a stratified one thousand-ticket test subset, versus zero point eight seven nine for fine-tuned DistilBERT on the same tickets. The findings are scoped to this dataset, prompt design, model endpoints, and illustrative costs; the work is an evaluation framework, not a new abstention algorithm or universal deployment certification.

Sources in this cluster