Over the past two years Espresso AI has built out a suite of observability tooling for customers of our automated cost-savings platform. Today, we're excited to announce that we're releasing this tooling for free to anyone with a Snowflake account.

Our Observability Philosophy

We've tried our best to curate the data we surface, aiming for actionable information instead of a barrage of charts and suggestions. Every chart and table on the dashboard started as an external user request, or an internal dashboard that we use to debug performance for us and our customers.

Features

A full overview of our features will be available on our docs in the coming days. Here are some highlights:

End-to-End Workload Latency

Snowflake warehouses have a ton of settings, but at the end of the day you're paying for just one thing: workload runtime. If you can pay less while hitting your SLAs, you should do that; the first step to understanding that is figuring out exactly what performance you're paying for.

This dashboard shows several metrics, the most meaningful one being your p99 latency. (Here's the classic article explaining why tail latencies matter if you're not used to tracking them.)

Query Cost Leaderboard

The most impactful thing you can do to bring down costs is understanding which queries are driving costs and how you can optimize them. Our leaderboard gives a daily breakdown of the most expensive queries, grouped by parametrized hash, including the query text, query plan, and runtime metrics.

We also include AI-suggested optimizations to help you start making improvements. Please thank our VCs for covering the token cost.

Bonus: if you tell people they're on the leaderboard they're going to try to get off of it.

Cost Breakdowns

Knowing which users, warehouses, and workloads are driving costs is the first step to bringing them down. This data is easy enough to pull directly from Snowflake, but it was also our #1 most-requested dashboard before we put it up.

Cluster Idle Time

Idle compute - the percent of time your Snowflake servers are sitting around doing nothing - is a core indicator of efficiency and workload health. Typical numbers here are 40-60%, with batch workloads seeing higher utilization and BI workloads seeing lower utilization. If yours is substantially higher, you should debug your settings.

(Our automated savings customers see idle time plummet when they turn on - leading to savings of 30-80% on compute bills.)

Storage Overview (Coming Soon)

Storage costs are typically 10% or less of the total Snowflake bill for most customers. If you're significantly above that and you've never looked into why, you're probably storing or backing up a ton of old, unused data. Our storage overview highlights total costs and

If you have a lot of old storage you can't remove, you should take a look at Snowflake's new storage lifecycle feature, or consider moving your storage to Iceberg and putting old files on Glacier.

‍

Contract Burndown

If you're on a long-term contract, our finance projection view will show you how your spend is tracking against your contract, including whether you'll need to renew early.

Get Started

Run the script here to get started - and don't hesitate to reach out with questions or feature requests.