How We're Using Snowflake Cortex To Deliver Self-Service Analytics

One month into a Snowflake Cortex pilot our Chief Actuary was convinced. He was speaking directly to the data and analytics team when he said, “Give this to us now, and we’ll stop bugging you with one-offs.”. Seeing was believing.

The Problem: Good SQL Isn’t Simple

Most organizations don’t struggle with too little data; they struggle with getting answers. When you start, you think it’s simple, but before long you come to realize that the data path is complex, and the right query even more so.

Enter Snowflake Cortex Analyst and Snowflake Intelligence.

The Breakthrough: Plain English Meets SQL

When we launched our Cortex Analyst pilot, we didn’t expect to see such a rapid shift. Within four weeks, business users were asking complex questions in plain English and getting accurate answers. Cortex Analyst handled nuance, industry jargon, and even the subtle phrasing of policy teams.

Better yet, it showed its work. Each time a user asked a question, Cortex Analyst displayed the SQL behind it. Transparency built trust. When it was right, users learned from it. When it was off, we’d retrain the model by correcting and re-running the data dictionary or the query. At this point, we are seeing increasing returns on our tweaks and tuning.

The Process: From Setup to Scale

We started with questions about policy data, relatively narrow questions. One person handled the setup by defining the data dictionary, building the semantic views, training Snowflake Cortex Analyst, and aligning with business users on sample questions. From there, the process became iterative and self-reinforcing.

This wasn’t a moonshot project. It was a low-risk, high-return way to experiment with AI on top of existing Snowflake data. For organizations already invested in Snowflake, Cortex Analyst offered an entry point into AI-driven analytics without new infrastructure or risk.

A tip from this observer: getting good results quickly needs three roles: a business SME, a Cortex SME, and a data SME.

The Impact: Security, Scale, and Speed

For the Chief Actuary’s team, the impact was immediate. They explored loss analytics across claim characteristics on their own. Questions that used to sit in backlog were answered in minutes. Every insight sparked the next.

Because Cortex Analyst runs entirely within Snowflake, data never leaves the secure environment. Role-based access, masking, and compliance policies all stay in place. Governance isn’t bypassed; it’s reinforced.

We dedicated a separate Snowflake warehouse to monitor Cortex Analysts' compute credit usage, allowing us to manage cost and performance proactively. With that in place, both teams could experiment freely and measure results.

The Vision: From Queries to Stories

Cortex Analyst solves a very specific part of a bigger journey—the point where users can finally ask questions and get answers. But that’s just one layer in the evolution of analytics.

The next step is AI-assisted BI, where tools don’t just return numbers but interpret them, compare them, and build a narrative around what they mean. Snowflake has more on the way to help with this next leg of the journey.  For now, we’re talking about getting answers, next comes insights.

Limitations

There are limitations that Cortex users must embrace.

First, you have to invest the time to create the data dictionaries with synonyms and table relationships, test questions, and train SQL. Some insurers are achieving scale by automating this step, piping data dictionary contents into Cortex to generate better results faster.

Second, there are limitations to how many tables a Cortex “analyst” can handle (about ten). That said, you can create views and nest Analysts to achieve domain-specific results. Expanding to enterprise scale is the next step on our roadmap.

Third, one may argue that Cortex provides answers but not insights; i.e., it does not interpret your data. More is coming from Snowflake on this front.

Snowflake Intelligence picks up where Cortex leaves off, allowing clients to mobilize agents that orchestrate bounded Cortex analysts.  This will enable users to move from asking what happened to asking why it happened.

The Takeaway: True Self-Service

True self-service isn’t about replacing analysts with AI. It’s about empowering business users to explore data confidently while maintaining governance and quality. Cortex Analyst proved that plain English-based data analysis is here and now.

This is plain English meets transparent SQL. Business and IT move in sync, and business moves faster. It doesn’t just answer questions faster; it bridges the distance between data and answers.