MEET ATSCALE / WHY A CATALOG ISN’T ENOUGH
Why a data catalog isn’t enough.
A catalog shows people where their data lives. AtScale computes the number your AI uses, so the AI doesn’t have to work it out from a description.
APPLES AND ORANGES
A catalog and AtScale do different jobs.
They’re built for different readers and they do different work.
A catalog helps people find and understand data.
A data catalog is an inventory of the data an enterprise owns: the tables and columns, who owns them, where they came from and what the business calls them. They’re built for people: a data steward uses one to document assets, and an analyst uses one to find the right table before writing a query.
After the search, it’s up to the analyst. They read the description, decide which columns to use, write the SQL and sanity-check the result against what they already know.
AtScale computes the answer your AI needs.
AtScale turns your business definitions into a semantic model that runs. It works out which metric a question means, computes the number in your warehouse and tests the answer. That way an AI assistant or an agent gets a number it can use, and nobody has to check it first.
“Catalogs are semantic helpers.”
Andrew Brust
Research Director, GigaOm, on an AtScale webinar
WHY THAT ISN’T ENOUGH FOR AI
AI reads the catalog and still has to guess.
An AI assistant or an agent can read the same catalog entry, but it doesn’t bring the analyst’s judgment. It turns the description into SQL on its own, and if it’s an agent, it acts on the result before anyone looks at it.
On BIRD-Interact, the public benchmark built from realistic enterprise questions, each model gets the schema, a description of every column and a knowledge base with every business formula. That’s roughly what a well-kept catalog offers. Even with all of it, an LLM on its own gets most business questions wrong, and adding AtScale makes the same model twice as accurate.
| Capability | A data catalog | AtScale |
|---|---|---|
| Built for | People searching for data | AI assistants, agents and BI tools asking for a number |
| Semantics | Descriptions, owners, lineage and tags | The same descriptions, owners, lineage and tags |
| Computing answers | Doesn’t compute | Computes each metric |
| When a person asks (prompt disambiguation) | Points to the tables that might answer it | Resolves which metric the question means, then computes it |
| When an agent asks | Points to the tables that might answer it | Gives the agent a protocol (MCP) to choose a precise, approved metric |
| Who writes the SQL | The analyst, or the LLM | AtScale ACE, as specified during semantic modeling |
| Proves the answer is right | No | AtScale Canary will test it |
| Keeps definitions under version control | Documents them | Stores them in Git, with review, tests and rollback |
HOW THEY WORK TOGETHER
Your catalog helps AtScale draft the model.
AtScale Semantic Modeling reads your catalog’s glossary, descriptions and lineage, and it uses them to draft your semantic model. Someone on your team approves each definition before it’s used.
HOW ATSCALE PROVES THE ANSWER
AtScale computes the answer and proves it.
Your catalog’s suggestions feed a semantic model that someone on your team approves, and ACE computes each answer in your warehouse. If an analyst asks a vague question, Ava checks which metric they mean before anything runs. If an agent calls the AtScale MCP server, it picks an approved metric by name and gets the answer back. Canary will test each answer against known correct results, and every miss goes back to the model as a fix.
Book a workshop.
- Bring us your most controversial metric and the business questions your team and your agents ask about it every week.
- We benchmark your AI on them, with AtScale and without it.
- You keep the semantic model we built to run the test.