Query Catalog
A query catalog defines your business once—what questions can start from, what numbers can be calculated, how results can be sliced, and what your team calls things.
When a question is covered by the catalog, Qluent doesn't write a query from scratch. It builds the answer from pieces you've already approved. While metrics get one calculation exactly right, a catalog covers a whole area of your data in any combination.
Think of it as a buffet
Everything on the table is there because someone put it there. You combine freely—any dish, any side. But you can't order something that isn't there.
That's how a catalog works:
- Only what's on the table gets picked - Nothing is improvised
- You decide what goes on the table - The catalog is yours to curate
- Combining is free - Any number, any slice, any period
- Nothing is faked to avoid saying no - A missing ingredient is a question, not a guess
Without a catalog, every question goes to a kitchen that improvises from a description. Usually good. Sometimes confidently wrong—and hard to spot, because a wrong number still looks like a number.
Why it's useful
- Same question, same answer - Every time, for everyone
- Your definitions always apply - "Orders" means successful orders, test accounts stay out, and nobody has to remember to say so
- No invented numbers - If something needed isn't there, you get a question back instead of a figure built on the wrong column
- You can see how an answer was built - Every result lists its source, numbers, groupings and filters
- Coverage multiplies - 6 metrics and 8 dimensions aren't 14 answers, they're every combination of the two—including questions nobody wrote down
What's in a catalog
Bases
The data a question can start from: orders, sessions, order items. Like the stations at a buffet—you start at one.
Each base carries its own rules:
- Which fields exist - Nothing outside that list can be used
- Default filters - Applied every time: successful orders only, no test accounts
- A default time window - Used when a question doesn't name a period (425 days), so open-ended questions don't scan all of history
- Your own named periods - Business vocabulary like "the final generated month"
Keep bases few: one per level of detail, not one per question.
Metrics
The numbers Qluent can calculate: order count, revenue, average order value.
- The recipe never changes, whatever the question
- Rates and averages are one metric divided by another, so the parts stay consistent with how they're calculated elsewhere
- Each metric is tied to the bases where it's really defined. "Order count" on an all-orders table isn't the same number as on a successful-orders table—some dishes are only served at certain stations
Your existing metrics are offered as catalog vocabulary, so you don't start from nothing.
Dimensions
The ways a result can be cut:
- Fields you already have - Country, channel, payment method
- Cuts you define once - Month buckets, size bands. "By month" then means the same thing in every answer
Dimensions are what make the buffet combinatorial. Each one multiplies what your existing metrics can answer.
Aliases
Nobody should have to learn your column names to ask a question.
- Word to field - "market" means
country_name - Word to value - "restaurants" means
restaurant - Word to cut - "monthly" means your month bucket
If a word could mean two available fields, Qluent asks instead of guessing. Aliases are the cheapest improvement to a catalog—when a question fails on wording, the fix is usually one line.
Example
User asks: "Top 10 markets by revenue last month, restaurants only"
| Part of the question | Comes from |
|---|---|
| "restaurants" | An alias, matched to the stored value |
| "revenue" | A metric, with your definition |
| "markets" | A dimension, via the alias for market |
| "last month" | A time window on the base's date field |
| "top 10" | A ranking on the result |
| unsaid, but applied | The base's default filters |
Swap revenue for order count, or markets for months, and it's a different question answered the same way —with no new configuration.
When the catalog doesn't cover a question
A catalog doesn't have to cover everything.
If a question falls outside it, Qluent asks a short clarifying question, or answers the way it normally does by writing a query against your data model. You still get an answer. You just don't get a catalog answer built on a guess.
So start small. Three well-covered areas are useful straight away, and failed queries show you where the gaps are.
Creating a catalog
- Go to Model → Catalog and select Generate query_catalog
- The assistant asks what your team measures—answer in plain language, you never write SQL
- Qluent drafts a catalog from your tables, descriptions, metrics and validated queries
- Review the draft, change anything you like, and save
Each save is a new version. Come back to the catalog whenever a wording fails, a new KPI matters, or your data changes.
Tips
- Curate deliberately - Fewer, well-defined items beat a long list nobody has reviewed
- One base per level of detail - A base per question gets hard to maintain
- Put standing rules in default filters - Don't rely on people phrasing questions carefully
- One metric per concept - Define rates as ratios of two metrics
- Use your team's words - Add aliases for the rest
The quality of a draft depends on your data model and instructions. Descriptions, sample values and validated queries all make it better.
Query catalog vs metrics
| Query Catalog | Metrics |
|---|---|
| A vocabulary Qluent builds queries from | A specific calculation |
| Covers an area of your data in any combination | Covers one KPI exactly |
| Bases, metrics, dimensions and aliases together | Name and SQL |
They work together: your metrics become part of the catalog's vocabulary.
To turn on catalog answering for a project, talk to your Qluent contact.