Natural Language to SQL: Qolaig x URW case study on Snowflake

Natural Language to SQL: how Qolaig made Unibail-Rodamco-Westfield's (URW) Snowflake data accessible to every team

‍
Introduction

According to a Sigma study reported in 2026, around 39% of business professionals aren't sure exactly what being "data-driven" means, and 76% of data experts say they spend half their time producing ad hoc reports for other teams. Forrester goes further: between 60% and 73% of enterprise data is reportedly never used for analysis at all. The paradox is striking: companies invest heavily in modern data warehouses like Snowflake, yet real access to that information stays concentrated in the hands of a few technical profiles.

That's exactly the observation that led Unibail-Rodamco-Westfield (URW) to bring in Qolaig to design an AI agent capable of turning a question asked in plain English into a reliable answer, drawn directly from its Snowflake data — with no SQL skills required on the user's end.

‍

Unibail-Rodamco-Westfield, a global leader in commercial real estate

Unibail-Rodamco-Westfield (URW) is a leading international commercial real estate group, owning and managing some of the most visited shopping centers in the world. A group of this scale naturally generates considerable volumes of data: footfall by center, occupancy rate, commercial performance by retailer, financial indicators by asset. This data was already centralized in Snowflake, one of the cloud data warehouses most widely used by large international groups — but centralizing data isn't enough to make it usable by everyone.

‍

The problem: rich Snowflake data, inaccessible to business teams

All of URW's data is hosted in Snowflake. It's available, but in practice inaccessible to the majority of employees. Getting an answer to a question as simple as "Which of our centers had the highest footfall this quarter?" meant either knowing SQL or asking the Data team — creating a bottleneck that slowed down decision-making at every level of the organization.

Ahead of the project, several key challenges were identified together with URW's teams:

• Enable business teams to query data with no technical skills required
• Reduce dependency on the Data team for routine queries
• Speed up decision cycles by removing intermediation delays
• Make Snowflake genuinely adopted and usable by non-technical profiles

This challenge isn't unique to URW: it's one of the most common friction points among large groups that have invested in a modern data warehouse (Snowflake, BigQuery, Redshift) without ever democratizing access to that information beyond their technical teams.

‍

Qolaig's solution: an AI Natural Language to SQL agent connected to Snowflake

Qolaig designed and deployed a Natural Language to SQL proof of concept (POC) for URW: a conversational interface that turns a question asked in plain language into a SQL query, executes it directly on Snowflake, and returns the answer in a clear, understandable way — without the user ever needing to know the underlying database structure.

‍

01. From question to data, no SQL required

The user asks their question the way they would ask a colleague, for example:

• "How many stores do we have in the United States?"
• "What's the average occupancy rate across our shopping centers?"
• "Which centers had the highest footfall this quarter?"

The system understands the intent behind the question, identifies the relevant tables in the Snowflake warehouse, generates the appropriate SQL query, runs it directly against the database, then interprets the results through a second AI layer before returning them in plain language — never exposing the user to the underlying technical complexity.

‍

02. A business glossary at the heart of reliability

The main difficulty in this type of project isn't technical, it's semantic. Every company has its own vocabulary: in-house metrics, internal synonyms, specific calculation rules, table relationships that are never fully documented. An AI Natural Language to SQL agent that ignores this business vocabulary produces queries that are syntactically correct but semantically wrong — a far more insidious risk than a simple technical error.

Qolaig built a custom business glossary for URW, allowing the AI agent to understand the exact definitions of the group's own metrics, the synonyms used by field teams, and the calculation rules specific to URW's real estate and commercial activity. It's this layer of business knowledge, built upfront with internal teams, that guarantees the relevance and reliability of the generated queries — far more than the raw performance of the language model used.

‍

Why Natural Language to SQL reaches far beyond URW's case

Natural Language to SQL (sometimes called text-to-SQL) fits into a broader logic of agentic automation applied to enterprise data: an AI agent that doesn't just answer a question, but goes and fetches the information directly from the organization's real information system, executes an action (here, a SQL query), and returns a usable result. Our comprehensive guide on agentic automation covers in more depth the perception, reasoning, and action mechanisms that make this type of agent possible.

This use case potentially applies to any company that has invested in a modern data warehouse (Snowflake, BigQuery, Redshift, Databricks) without solving the access problem for non-technical profiles: senior management, sales teams, marketing, operations. The bottleneck observed at URW — a Data team overwhelmed with routine queries, an intermediation delay that slows down decisions — shows up identically in finance, insurance, retail, industry, or logistics, sectors we cover in detail in our guide on AI agents in the enterprise.

‍

The results achieved

• Natural-language data access for all business teams, with no technical skills required
• Reduced workload on the Data team for routine queries
• Easier Snowflake adoption by non-technical profiles, who previously only had indirect access to the data
• Faster decision-making thanks to the removal of intermediation delays between a business question and its answer

‍

Return on investment

Every question that made its way to the Data team consumed skilled time on low-value tasks — exactly the kind of burden revealed by the Sigma study cited in the introduction, with data experts spending a significant share of their time on ad hoc reporting rather than higher-impact work. By delegating these routine queries to an AI agent, URW frees up its data engineers and data analysts for higher-value work: advanced modeling, strategic projects, data governance — all while speeding up decision cycles for the business teams themselves.

For a more general method of calculating the ROI of an AI agent project, our guide on the price of an AI agent in the enterprise covers the levers to factor in: time freed up, error rate reduction, faster decision cycles.

‍

FAQ

What is Natural Language to SQL (text-to-SQL)?

It's a technology that turns a question asked in natural language into an executable SQL query against a database, without the user needing to know SQL or the structure of the database being queried.

‍

Is it secure to connect an AI agent directly to a data warehouse like Snowflake?

A well-scoped project defines a clear access perimeter, sets clear governance rules on which data can be queried, and builds in human oversight for sensitive use cases — security and data governance need to be designed into the project from the start, not bolted on afterward.

‍

What's the difference between a Natural Language to SQL AI agent and a traditional Business Intelligence tool?

A traditional BI tool generally requires building dashboards and reports in advance, around anticipated questions. A Natural Language to SQL AI agent answers freely formulated questions, including ones that weren't anticipated in advance, by generating the corresponding query on the fly.

‍

Why is a business glossary essential for this type of project?

Because the main source of error isn't technical but semantic: without understanding a company's vocabulary, synonyms, and calculation rules, an AI agent can generate a technically valid query whose result doesn't actually match what the user was really looking for.

‍

Is this type of project reserved for large groups with a data warehouse like Snowflake?

No. The same logic applies to any company with structured data, regardless of the tool used to host it. What matters isn't company size, but the volume of recurring questions that currently get escalated to a technical team when they could be handled independently by business teams.

‍

Take action

This project illustrates a core conviction of our approach at Qolaig: the value of an AI agent doesn't rest solely on the performance of the model used, but just as much on a deep understanding of the business context in which it operates. If your organization holds rich data that remains out of reach for non-technical teams, that's exactly the kind of problem our Workshop → Build → Run methodology is designed to scope and solve. Let's talk about your project.