Customer service
Which models had the most warranty repairs?
A question that used to need an export and a spreadsheet now gets a read-only answer straight from the ERP database, with the area's rules already applied.
Module 01/06 · NOEVA Analytics
An AI that answers new questions about your data. And when it doesn't know, it says so.
01 The problem
02 The idea
The hard part isn't writing the query. It's knowing what the data means.
03 What the data means
To sales, an “active customer” ordered in the last year. To accounting, they have an open invoice. To customer service, they own a product under warranty. Every area has its own vocabulary, and it has to be written down.
04 One workspace per area
NOEVA Analytics works by project. Each area of the company has its own data source, its own vocabulary and its own authorised people.
05 The traps, written down once
A chain registered under two customer codes. A return that isn't a complaint. We find them together and write them down once: from then on they apply to every question.
06 From question to query, read-only
Service asks which models had the most warranty repairs. The agent understands the question, applies the area's rules, writes the query and runs it read-only, on your database.
07 Better an explicit error
If a join between tables isn't declared, the query doesn't run. Better an explicit error than a plausible, wrong number.
08 The same context, for every agent
Each workspace's context is available to authorised agents through the MCP standard. Data never leaves the company, and everyone sees only their own workspace's projects.
09 Use cases
Customer service
A question that used to need an export and a spreadsheet now gets a read-only answer straight from the ERP database, with the area's rules already applied.
10 Where to start
Pick one area of the company.
Pick the report that costs you the most time.
In six weeks we rebuild it with NOEVA Analytics.
If the numbers don't match, we stop there.
11 Frequently asked questions
No. Queries run read-only on your database.
The agent applies the area's vocabulary. If a link between data isn't declared, the query doesn't run and it says so, instead of making up a number.
No. Context is available only to authorised agents, via MCP, and everyone sees only their own workspace.