Ask your database in plain language and get a table or a chart back
A manager types a question in plain words: “sales by store this week compared to last week”. The system writes a SQL query against your database, returns a table or chart, and shows the query so it can be checked.
What hurts
Every new question about the data goes through an analyst or a developer. The report queue stretches over days. By the time numbers arrive, the decision has already been made on gut feeling.
You have dashboards, but they answer only the questions someone thought of in advance. Slice the data a new way and you need an analyst again. Executives stop opening the BI tool.
Letting staff query the database directly is risky. One bad query can load production or expose data people should not see. So data access stays with a narrow circle.
What we build
- Data model description: tables, joins, business terms and metric rules
- Connection to PostgreSQL, MySQL, BigQuery or your data warehouse
- SQL generation with validation before execution
- Read-only access with limits on query time and result size
- Answers as a table, chart and text, with the SQL shown
- Table and field restrictions by user role
- Web interface or a Telegram bot
- Test set of business questions with verified answers
What you get
- A working data query service with logins for your team
- An editable semantic layer describing metrics and terms
- Question test set and an accuracy report
- Source code, access and documentation
- Training for users and for whoever maintains the semantic layer
How we measure the result
- Share of test questions answered with the correct figure
- Questions per week from managers without an analyst involved
- Time from question to answer
- Number of queries rejected as unsafe or ambiguous
Systems and stack
Industries
Questions
How much does text-to-SQL cost?
From $7,000. The price depends on the number of tables and sources, how complex the metric rules are and the access requirements. Most of the time goes into defining metrics with your team. We fix the estimate within 2-3 days after discovery.
Can it break the database?
No. We connect with a read-only user, ideally to a replica or a warehouse. Every query is validated before it runs and has a time limit. The system technically cannot change or delete data.
How accurate are the answers?
AI sometimes misreads a question, especially an ambiguous one. So the system shows the SQL and explains how it calculated, and asks back when a question is unclear. We check accuracy on a business question test set before launch and after changes.
Which systems does it work with?
PostgreSQL, MySQL, BigQuery, ClickHouse and data warehouses. Data from SAP, Odoo or a CRM is first consolidated into one store if that is not done yet. We can do it within the project.
What do you need from us?
Read access to the database or a replica, a list of typical management questions and a person who knows how your metrics are calculated. We sign an NDA and a DPA before any access.
Related services
More in this section
Describe the task. We reply within one business day, the first call is with the founder.
Describe the task. The first call is with the founder.
We reply within one business day. 30 minutes to tell you whether it is worth doing and what it costs.