The warehouse answers business questions in conversation
Business users previously accessed consolidated commerce data only through prepared reports. Every other question became a request and had to wait.
Four tools now lead from business documentation through table listings and schema descriptions to read-only queries. The customer retains its own AI application.
The starting point: indirect access to answers
Questions from purchasing, controlling and product management cut across stock, sales, margins and marketplaces. No prepared report covered that combined view.
The turning point: four tools with a defined access boundary
Consolidated metrics
Reporting tables combine revenue, gross profit, stock, stock cover and ABC classes. One question can now query them together.
Secured the endpoint
During implementation, we found that the database timeout did not cover certain queries. We added a second service-level deadline that forcibly closes hanging connections. SQL validation permits only individual read queries, with row and time limits.
Managed access
Tokens are created in the application, which stores only a hash. Every query carries the client name, and deleting a token invalidates it immediately.
Documented the data for AI
Curated documentation explains 32 tables, joins and query recipes so product, stock cover and gross profit have consistent meanings.
McFilter is a medium-sized e-commerce company. Purchasing, product development, warehousing and shipping are handled in-house, while sales run through marketplaces and the company’s own shops.
