🔍 Cost-Ops in Snowflake: 7 Practices That Slashed Our Bill by 30 %

While working on an internal project—Nimbus Intelligence Data Warehouse—I was mainly responsible for cost-optimization settings in Snowflake so that my team and I wouldn’t exceed our budget. After several days of research, I identified seven key actions that I implemented on the project and that could cut your Snowflake bill by as much as 30 percent if you apply them, too.

🗺️ Visual first: how does every piece fit together?

graph TD
%% — Observabilidad —
subgraph Observabilidad
A[WAREHOUSE_EVENTS_HISTORY] –>|uso real| B[QUERY_ATTRIBUTION_HISTORY] end

%% — Gobernanza —
subgraph Gobernanza
C[Tags & Budgets] D[Resource Monitors] end

%% — Automatización —
E[Triggered Tasks] –> F[Auto-downscale]

%% — Relaciones cruzadas —
A –> D
B –> C
C –>|Alertas / Webhooks| Fin(FinOps)

Paste this snippet into mermaid.live or export a PNG for your cost-committee deck.
ACCOUNT_USAGE tables (blue boxes) are the radar; monitors and budgets (red) are the sirens; triggered tasks fire the corrective actions.

1️⃣ Right-size and auto-suspend your warehouses

• Aim for the “just-enough” size: start at S and scale only if WAREHOUSE_LOAD_HISTORY > 80 %.
• Enable AUTO_SUSPEND = 60 and AUTO_RESUME = TRUE.
• Inspect WAREHOUSE_EVENTS_HISTORY to hunt for ghost resumes.

2️⃣ Resource Monitors—cut the card before it melts

Create one monitor per environment and set thresholds:

CREATE RESOURCE MONITOR rm_dev WITH CREDIT_QUOTA = 1000 
  TRIGGERS ON 80 PERCENT DO NOTIFY
           ON 100 PERCENT DO SUSPEND;

Your warehouse pauses before the budget is blown.

3️⃣ Set Budgets and keep FinOps in the loop

Budgets provide monthly limits and predictive alerts (webhook or email). Use them for serverless spend (Snowpipe, Tasks) that monitors don’t cover.

4️⃣ Attribute spend with Tags + Query Attribution

Tag databases, schemas and warehouses:

ALTER WAREHOUSE etl_wh SET TAG cost_center = 'marketing';

Then query QUERY_ATTRIBUTION_HISTORY to know who spent how much, and why.

5️⃣ Run only when data is present with Triggered Tasks

You used to fire a cron every 5 min; now trigger when a micro-batch arrives:

CREATE OR REPLACE TASK t_load_orders
  WAREHOUSE = ingest_wh
  SCHEDULE = 'USING CRON 0 * * * * UTC' -- “old school”  
  WHEN SYSTEM$STREAM_HAS_DATA('stg_orders_stream')
AS CALL load_orders();

With Triggered Tasks, the WHEN … clause replaces the cron: zero idle time, zero wasted credits.

6️⃣ Shamelessly exploit Zero-Copy Cloning

Clone 10 TB databases in seconds without paying storage; you’re charged only for changed blocks. Perfect for dev/QA. Drop stale clones so they don’t inherit the parent’s costs.

7️⃣ Put a FinOps bot in production

Combine:

• A stream on WAREHOUSE_EVENTS_HISTORY
• A triggered task that downsizes or suspends
• A Slack webhook + Jira ticket

Turn cost post-mortems into preventive actions with zero human intervention.

✅ Checklist for tomorrow

1. Review warehouse sizes and AUTO_SUSPEND.
2. Create Resource Monitors per environment.
3. Enable Budgets at account level + one per project.
4. Tag critical objects and enable QUERY_ATTRIBUTION_HISTORY.
5. Migrate cron tasks to Triggered Tasks.
6. Document a Zero-Copy Cloning policy (14-day TTL).
7. Fire a Slack alert when usage > 80 % of budget.

💬 Your turn

Which trick saved you the most Snowflake credits?