Snowflake warehouse optimization with Slingshot

As the data landscape is constantly evolving, data and platform teams need a centralized system to manage, govern and optimize their data clouds at scale. At enterprise-scale, these systems and your pipelines can quickly become riddled with inefficiencies, driving unnecessary spend.

Capital One Slingshot was built out of a need to manage and optimize Capital One’s own data infrastructure. Facing the complexities of managing Snowflake at an enterprise-scale, our internal teams developed Slingshot to help manage and optimize petabytes of data in this system. 

What began as a powerful internal tool evolved into a commercial software offering available to Snowflake customers. Today, Slingshot empowers data-intensive companies across various industries to achieve the same level of granular control, optimization and visibility over their data cloud spending and pipeline performance.

In this post, we lay out how Slingshot’s dynamic warehouse schedules for Snowflake can help you cut costs and optimize with confidence, while maintaining full control over your Snowflake environment. Let’s dive in!

Snowflake warehouses come in t-shirt sizes that range from XS up to 6XL. Warehouses also have various options for cluster counts and configuration settings (Warehouse Type, Auto-Suspend, Auto-Resume, etc.), for users to pick from. As a result, there are many different configuration options that can directly impact the cost and performance of a warehouse. 

The table below shows Snowflake’s Gen1 warehouse sizes, the corresponding credit consumption and maximum allowable cluster counts per t-shirt size:

* Sources available from Snowflake documentation: Multi-cluster warehouses and overview of warehouses.

A warehouse size directly impacts its cost and performance. Additionally, horizontal scaling, using multi-cluster warehouses (with up to 300 clusters for the smaller warehouse sizes), helps with concurrent workloads, which can also have a direct impact on credit consumption.

Lastly, Snowflake’s newer Gen2 warehouses come with additional considerations. Gen2 warehouses consume more credits per hour than their Gen1 counterparts, but their efficiency gains can compensate for the premium rate, especially in DML-heavy environments. In fact, Slingshot examined more than 2 billion query profiles across +14K warehouses in April 2025 and found up to 25% cost savings for DML operations on Gen2 vs their Gen1 counterparts. 

Slingshot helps organizations optimize their Snowflake virtual warehouses with dynamic schedules for both Gen1 and Gen2 warehouses. Schedules go down to the hourly level and are based on the warehouse’s actual performance and usage. You can easily migrate a warehouse from Gen1 to Gen2 in Slingshot and understand the impact of that change using the reports in the UI. Think of this as a low-risk way of testing the new Snowflake compute.

With all of the options available to Snowflake users, there can be a lot of uncertainty around ensuring workloads and pipelines are operating efficiently.

Common questions we hear from customers include:

• How do I know that my warehouses are sized correctly?
• How utilized are my warehouses?
• Are my workloads running efficiently?
• How do I know if my warehouses are optimized?

With Slingshot, dynamic schedules rightsize your warehouses at scale. Each schedule is tailored to that warehouse based on its cost and performance. By analyzing historical performance data, Slingshot provides data-driven recommendations based on real usage patterns. This analysis spans key data metrics such as query load, queuing, idle time, query size and complexity and data spillage.
 

Using this data, Slingshot identifies warehouse issues, such as low/high query load, low/high spillage or no auto-suspend set. It then recommends a dynamic schedule to optimize the warehouse. For more on this, read this post.

Slingshot’s warehouse schedules include warehouse size changes, cluster minimum and maximums and auto-suspend thresholds. Each recommended schedule also provides cost and query impact projections, so that users can understand the implications of a schedule on warehouse cost and performance before they decide to apply it.

Recommendations include gradual changes, with each recommendation building off of the cost and performance impact of the previous one.

For example, if an underutilized XL warehouse generally has low data spillage, runs primarily short queries and has a low query load during off-peak hours, Slingshot would recommend that warehouse to downsize to a size L for those timeslots. The next recommendation cycle may find that the warehouse is still underutilized during off-peak hours, and suggest the warehouse be downsized to a M during those hours. These recommendations will continue to update based on warehouse workload and demand, and allow customers to walk the line between cost and performance at scale.

Slingshot also comes with governance features like fine-grained RBAC, warehouse templates and approval workflows. If we take the dynamic schedule recommendation example, a new schedule can be submitted to the warehouse owner for review. The owner can review the request and make any necessary adjustments (if required) before the schedule is applied. These governance features can also be applied to broader warehouse provisioning and management. Snowflake admins can create warehouse templates with built-in guardrails, so that the team can provision the desired warehouses independently.

Slingshot’s data-driven optimizations are supported by detailed impact analyses on applied recommendations. The warehouse details page provides a unified view to review warehouse performance over time. It offers granular insight into a warehouse's historical cost and performance trends, resource utilization and calls out when changes were made to its configuration.

You can see detailed information on existing configuration settings and cost and performance changes as warehouse configurations change. This provides a clear view of warehouse cost and performance, along with the ability to assess the impact that each optimization has had on those metrics. 

Slingshot dynamic schedules automatically rightsize your warehouse, but you retain full control. You can implement single or multiple-step approval workflows to protect your mission-critical workloads. Warehouse owners will review the recommendation and tweak it if needed to fit specific requirements, before it can be applied to the warehouse. Once reviewed, the recommendation is applied and the dynamic schedule is automatically put into place without ongoing manual effort. Alternatively, you can enable relevant team members to apply optimizations to certain warehouses, without the owner’s approval, using resource-level RBAC.

Regardless of how you choose to apply recommendations, you have the ability to edit any dynamic warehouse schedule or revert back to a previous configuration at any time.

Slingshot’s dynamic warehouse schedules for Snowflake help you rightsize your warehouses at the hourly-level without the manual lift. This helps organizations find a balance between cost and performance optimization.

With dynamic schedules, Slingshot automatically rightsizes your warehouses based on real-world usage, helping you prevent unnecessary spend.

Every change and recommendation applied is automatically logged and flagged in the UI so you can easily assess the impact of each change on costs and performance. The benefit here is twofold: 

1. You can validate the impact of changes made on warehouse cost and performance.
2. You can revert back to a previous configuration, if the impact is not as intended.

Finally, Slingshot’s governance features spanning resource-level RBAC, approval workflows and warehouse templates help platform administrators maintain policy and control, while unblocking the team to self-serve.