AI on company data fails without one definition per metric
In brief
According to Qonto, the neobank first turned business questions into SQL, then let Claude query the data warehouse. Different metric definitions across teams led to incorrect answers. The company then built a catalogue of approved metrics and dimensions; it targets coverage above 90 percent and, by the end of 2026, all priority metrics and questions.
Source : Silicon.fr - « Comment Qonto a basculé du text-to-SQL à l’analyse agentique »
What it changes for an SME
For an SME with 10 to 250 staff, a natural-language assistant on your data can return the wrong numbers if each team defines metrics differently. According to Qonto, engineering spent several months on existing models, with no new features delivered, and other projects were postponed. Business users then have to judge the results themselves, because an analyst is no longer in the middle. Cost shows up as engineering time; risk shows up as poor decisions.
What you can do
- Check that each important business metric has a single official definition.
- Run a few key questions through an AI data tool and compare the answers with figures your teams already trust.
- Set aside engineering time to clarify data models before you open natural-language access.
Let's talk about the time you could save.
In 20 minutes, we find where AI could save you time or money. Free, no commitment.