World Journal of Engineering Research and Technology

*Anil Kumar Kolla, Ameya Kokate

ABSTRACT

Natural-language interfaces to enterprise data promise to remove the analyst as a bottleneck between a business question and its answer. Large language models have made such interfaces practically deployable, and major data platforms now ship conversational query capabilities over governed enterprise datasets. However, the evidence base for their enterprise use remains thin: published text-to-SQL benchmarks measure execution accuracy against curated academic schemas, whereas enterprise deployment is constrained by ambiguous business vocabulary, undocumented schemas, row- and column-level access control, and the practical consequence that a confidently wrong answer delivered to a decision-maker is worse than no answer at all. his paper contributes an evaluation framework for natural-language analytics in enterprise settings. We define a taxonomy of enterprise query classes distinguishing lookup, filtered aggregation, comparative, temporal, and metric-definitional queries; we specify a metric suite extending execution accuracy with semantic-equivalence accuracy, abstention rate, harmful-answer rate, and governance conformance; and we describe a semantic-preparation methodology — curated question-answering scope, certified metric definitions, business glossary binding, and worked example pairs — that treats natural-language readiness as a data product responsibility rather than a model capability.

[Full Text Article] [Download Certificate] https://doi.org/10.5281/zenodo.23120385