A natural-language query is a question put to a data system in plain words instead of code: what did we pay per tonne for pine sawlogs last quarter, by supplier? An AI model translates it into a database query, usually SQL, runs it and returns the answer. It is one kind of what Wikipedia calls a natural-language user interface, "where linguistic phenomena such as verbs, phrases and clauses act as UI controls for creating, selecting and modifying data".
How well it works
It depends on the data. The authors of BEAVER, a benchmark built from private enterprise data warehouses, note that most earlier benchmarks use "public databases with well-structured schemas", where models do well. BEAVER holds 9,128 questions and queries across 812 tables, built from real query logs with expert-verified additions. On it, leading agentic systems "achieve only 10.8% accuracy". Given expert hints for each step, accuracy "increases to 30.1%".
That is one-shot accuracy on hard analytical questions over unfamiliar warehouses, with no written definitions. A system set up over a curated model with agreed definitions is in a different position.
Much of the gap is domain knowledge. Company databases have many tables, cryptic column names and definitions that exist only in people's heads. Which volume, gross or net? Which cost, with or without freight? A model guessing those answers produces a query that runs, returns a number and is wrong.
In a timber business
Timber data adds units and identity changes: tonnes and board feet, tickets and tallies, stands and contracts. A plain question about cost per unit can be answered several defensible ways, and the model needs to be told which one the business uses.
What makes it work
Agreed definitions, written where the system can read them, sometimes called a semantic layer. Known joins between systems. Answers that show the tables, filters and definitions used, so a person can check them.
What to check
Test with questions whose answers you already know, including awkward cases such as credits and transfers. Ask to see the query behind each answer, and check that the system says when it can't answer.
How Quarri works explains the platform as a layer over existing systems, not a migration.
Sources
- Wikipedia, "Natural-language user interface": en.wikipedia.org
- Chen and others, "BEAVER: An Enterprise Benchmark for Text-to-SQL", arXiv 2409.02038, revised 13 May 2026: arxiv.org
Quarri is an AI-native data platform for the timber supply chain. It connects buying, production, sales and inventory for forest management, sawmill, wood products and pulp, paper and packaging operators.