Analytics context

FeatureQL features are pure functions — they describe what to compute, not where the data lives or which rows to process. Batch analytics is where you connect those features to real data and evaluate them across entire entity populations.

From single-entity to population-scale

In real-time serving, you evaluate features for one entity at a time: "What is the lifetime value of customer 42?" Batch analytics flips this: "What is the lifetime value of every customer who was active last month?"

The building blocks for this are:

• EXTERNAL_COLUMNS() and EXTERNAL_SQL() — connect features to columns in your data warehouse tables
• @BIND_KEYSET() — define which entity keys to evaluate, often filtered by date or business criteria
• GROUP BY and UNNEST() — aggregate results or flatten nested structures for reporting
• Window functions — compute running totals, rankings, and row-level analytics with OVER() clauses
• Hybrid queries — embed FeatureQL inside SQL (or vice versa) for seamless integration with existing tools

Query structure

A batch analytics query follows this general shape:

WITH
    <feature definitions>
SELECT
    <features to compute>
FOR
    <bind entity keys>
WHERE <filter>
GROUP BY <aggregate>
ORDER BY <sort>
LIMIT <rows>

WITH
    <feature definitions>
SELECT
    <features to compute>
FOR
    <bind entity keys>
WHERE <filter>
GROUP BY <aggregate>
ORDER BY <sort>
LIMIT <rows>

The WITH and SELECT clauses define features. The FOR clause binds entity keys — this is what makes the query evaluate across a population rather than a single entity. The remaining clauses (WHERE, GROUP BY, ORDER BY, LIMIT) shape the output.

What's in this section

Each page covers one piece of the batch analytics workflow: