North America – October 27, 2026 -- Snowflake's product data science team cut the time from pipeline incident to a tested, reviewable fix to under 16 minutes using its CoCo agentic AI assistant. Before CoCo, diagnosis took a skilled engineer about 30 minutes per case. The result comes from a four-part framework that Snowflake built to measure AI value beyond license counts and prompt volumes.
Snowflake sorts AI work into three levels instead of counting prompts
Dashboards built on access and activity score an organization running 1,000 explanations the same as one running 1,000 pipeline builds and anomaly investigations. Snowflake's framework separates foundational work (explaining code, answering questions), applied development (writing code, exploring data, building reports) and agentic work (multistep processes where AI uses tools to finish an outcome). Real capacity gains should appear in the agentic tier.
The framework tracks adoption (breadth across the eligible population) separately from fluency (sophistication of delegated work), then benchmarks both against industry peers. A software company and a regulated insurer do not share a baseline, so a 35% adoption rate only means something relative to comparable organizations. The fourth step studies the gap between top users and everyone else.
Product data science automated 160 hours a month of diagnosis
In the first quarter of FY 27, the team's pipelines needed attention 960 times. At about 30 minutes per case, that is roughly 160 hours a month of expert diagnosis. The team handed its existing diagnostic runbook, refined over hundreds of incidents, to CoCo and set an alert to trigger it.
Across Q1 and Q2, CoCo improved more than 500 pipelines, generated more than 100 merged pull requests and reduced hundreds of daily wasted credits. Snowflake notes that optimizing expensive pipelines is valuable but rarely urgent, so the work tends to wait indefinitely without an agent.
CoCo only proposes changes. Each arrives as a pull request that a named owner reviews and approves, while merging and deploying stay with the team. Snowflake warns that reviewer attention becomes the new constraint if a team generates more proposals than it can read.
Marketing weekly CoCo use rose from eight people to 70% of the team
Snowflake's marketing organization has nearly 600 people. A February 2025 internal survey found 11.6% of the North America marketing team felt very confident using AI in their role. When CoCo launched as a research preview late that year, eight North America marketers used it weekly. By May 2026, weekly CoCo usage across global marketing had reached 70%. Separately, 93% said they use AI tools daily, a self-reported figure that Snowflake ranks below observed behavior.
Marketers built tools that replaced manual work:
- A motion graphics export pipeline that compressed a full week of Summit work into 20 minutes.
- A win-loss reporting app on CoCo and Streamlit, now roughly 90% automated, replacing three hours a month of candidate identification plus a week of deck work.
- An inbound automation engine handling 2,000 to 3,000 daily email replies to global campaigns with no human intervention, replacing a planned software purchase.
Marketers have created more than 600 personal and project-level skills, and 100 to 200 built their own from scratch. The median user runs seven skills a week. Snowflake credits quarterly AI Days, regional hackathons, a newsletter and office hours, all peer-led.
Three usage profiles call for three different responses
For organizations with sophisticated but narrow usage, Snowflake recommends removing access friction and expanding enablement. Broad but shallow usage calls for templates and hands-on training that show how to delegate larger tasks. Where usage is both broad and sophisticated, adoption stops being the priority and the focus shifts to reusable agents and outcomes such as development velocity or operational efficiency.