LO
DeploymentCognitive automation2025-07-18
LinkedIn says its text-to-SQL bot has had over 300 weekly users since July 2024, that dataset owners' annotations lift correct answers from 9% to 48%, and wrong filters are the top failure
Data analystoccupation page →
Event date / reported
2025-07-18
Evidence stage
DeploymentAn employer has put it into production. Can move the baseline — weighted by scale and how similar the setting is.
Tasks this bears on
Working out what is really being asked
Turning a manager's vague worry into a measurable question, and knowing which definition of 'active user' they mean.
Still human-led✓ Evidence-backed
Owning the definitions the tools rely on
Maintaining the metric definitions, the documentation and the semantic layer that make natural-language querying produce correct answers.
New task✓ Evidence-backed
Where this applies
A paper by LinkedIn staff about their internal text-to-SQL chatbot, available in LinkedIn's querying platform since July 2024 with steady use by over 300 weekly active users. It relies on a knowledge graph in which dataset owners annotate whether a column is a metric, dimension or attribute and popular tables have human-authored descriptions; these components raised answers scoring as correct or nearly correct from 9% to 48%. In an expert review of 124 production answers, 53% were rated correct or with minor issues, and the most common issue was an incorrect filter (24%); the authors note that a request for the latest click-through rate means different things to different teams. It is one company's internal tool, measured by its builders.
What this means
Natural-language querying works at one large company only as well as the definitions people maintain behind it — owners' annotations took correct answers from under a tenth to about half — and its commonest error is getting the question's filter wrong. Both the definitions and the framing stay with analysts.
What it does not yet show
One company's tool measured by its builders; it does not show fewer analysts or how common such tools are.
What you can check
Open arXiv:2507.14372 and find "Column type indicates whether a column is a metric, dimension, or attribute, as annotated by the dataset owner".
Does it change the assessment?
No. The impact index is never moved by a single event. Of the 2 linked judgements above, 1 moved from inference to evidence with this record; the other 1 already rested on earlier evidence.
Source
Chen et al. (LinkedIn) — "Text-to-SQL for Enterprise Data Analytics", arXiv:2507.14372 (18 July 2025; Workshop on Agentic AI for Enterprise at KDD '25) · verified 2026-09-27 · Claude (VOLO agent) · interpreted 2026-09-27 · Claude (VOLO agent)
Primary source — published by the party that did this, or the authority of record. No co-signature needed.