Snapshots — dbt's SCD Type 2

Snapshots are dbt's built-in Type 2 SCD tool.

The snapshot block

-- snapshots/customers_snapshot.sql
{% snapshot customers_snapshot %}

{{
    config(
      target_schema='snapshots',
      unique_key='customer_id',
      strategy='timestamp',
      updated_at='updated_at',
    )
}}

SELECT id AS customer_id, email, country, segment, updated_at
FROM {{ source('shopify', 'customers') }}

{% endsnapshot %}

Run dbt snapshot. First run creates the table with dbt_valid_from, dbt_valid_to, dbt_scd_id. Subsequent runs detect changes, expire old rows, insert new versions.

Two strategies

timestamp

Uses source's updated_at. Row is "changed" if updated_at is newer.

strategy='timestamp', updated_at='updated_at',

check

Compares specific columns row-by-row.

strategy='check', check_cols=['email', 'country', 'segment'],

Use when source has no reliable updated_at.

Columns dbt adds

• dbt_scd_id — surrogate key for this version.
• dbt_valid_from — when this version became active.
• dbt_valid_to — when superseded (NULL = current).
• dbt_updated_at — source updated_at value.

Querying

Current state:

SELECT * FROM {{ ref('customers_snapshot') }} WHERE dbt_valid_to IS NULL

State at a specific time:

SELECT * FROM {{ ref('customers_snapshot') }}
WHERE customer_id = 7
  AND dbt_valid_from <= '2026-04-01'
  AND (dbt_valid_to > '2026-04-01' OR dbt_valid_to IS NULL)

Joining facts to snapshots

The point — facts join to the version active at fact time:

SELECT o.order_id, c.country, c.segment
FROM {{ ref('fct_orders') }} o
LEFT JOIN {{ ref('customers_snapshot') }} c
    ON c.customer_id = o.customer_id
    AND o.order_placed_at >= c.dbt_valid_from
    AND (o.order_placed_at < c.dbt_valid_to OR c.dbt_valid_to IS NULL)

Where to put snapshots

Separate snapshots/ folder. Separate schema (snapshots or dbt_snapshots). Snapshots accumulate history forever and shouldn't be rebuilt.

Common mistakes

1. Skipping snapshots, using table for dims — lose history.
2. Running infrequently for volatile data — miss intermediate states.
3. Including high-velocity columns in check_cols — every login generates a row.
4. Querying WHERE dbt_valid_to IS NULL everywhere — defeats Type 2.

Takeaway

Snapshots = dbt's Type 2 SCD. Pick strategy, point at source, schedule. Join on time-range conditions for history-correct queries.