Snowflake Cortex AI Explained: Capabilities, Pricing, and Real-World Use Cases

Snowflake Cortex AI transforms how organizations leverage artificial intelligence by bringing powerful AI capabilities directly into the Snowflake Data Cloud. Rather than moving data to external AI services, Cortex AI enables you to apply large language models, search, and analytics directly where your data lives, secure, governed, and integrated.

Part of Snowflake's broader AI strategy, Cortex AI provides a comprehensive suite of managed services: from LLM functions for text processing to specialized tools like Cortex Analyst for natural language querying and Cortex Search for intelligent retrieval.

In this article, we'll explore the Cortex AI suite and demonstrate one of its newest capabilities: AI_REDACT for automatically detecting and removing personally identifiable information (PII) from unstructured text.

What is Snowflake Cortex AI?

Cortex AI transforms Snowflake from a data platform into an AI-native execution environment. Rather than treating AI as a separate stack, Cortex integrates intelligence directly into your data workflows, eliminating the operational, security, and compliance challenges that traditionally accompany enterprise AI adoption.

• Core principle: AI where your data lives.

Cortex AI operates within Snowflake's security perimeter, respecting all existing governance policies, RBAC controls, and data residency requirements while providing access to industry-leading models from Google, OpenAI, Anthropic, Meta, Mistral, and more.

The Cortex AI Suite

Cortex AI consists of five integrated components:

1. Snowflake Intelligence: Unified conversational AI interface enabling business users to analyze and act on data using natural language without SQL. Example: A sales manager asks, "Show me Q4 revenue by region," and gets instant visualizations without writing queries.

2. Cortex Agents: Orchestrate tasks across structured and unstructured data by coordinating between Cortex Analyst, Cortex Search, and LLMs. Example: An agent automatically retrieves customer contracts, analyzes sales data, and generates renewal recommendations.

3. Cortex Analyst: Fully managed service that converts natural language questions into accurate SQL queries using semantic models. Example: "Which products had declining sales last quarter?" becomes optimized SQL against your data warehouse.

4. Cortex Search: Managed RAG service combining semantic vector search with keyword matching for AI chatbots and document search. Example: Search across millions of support tickets to find similar issues and their resolutions.

5. Cortex LLM Functions: SQL and Python functions providing direct access to large language models for text generation, classification, extraction, and analysis. Use AI_SUMMARIZE to condense lengthy documents or AI_EXTRACT to pull key entities from unstructured text.

How Cortex Code and Cortex AI Work Together

Cortex AI is most powerful when paired with Cortex Code. While Cortex AI provides SQL-based intelligence functions, summarization, classification, redaction, sentiment analysis, and semantic search, it assumes that data pipelines, schemas, and governance structures are already in place. This is where Cortex Code plays a foundational role.

Cortex Code accelerates:

-Data model creation

-dbt project generation

-Governance validation

-Workflow orchestration

-Application scaffolding

Once that structure is built, Cortex AI functions can safely operate on curated datasets without exposing raw or ungoverned data.

In enterprise environments, this separation matters because Cortex Code ensures architectural discipline, and Cortex AI enables intelligent execution. Together, they support a secure, scalable AI adoption model inside Snowflake, from infrastructure to inference.

Cortex AI in Action: Privacy-Safe Analytics with AI_REDACT

Let's demonstrate one of Cortex AI's newest capabilities: automatically detecting and redacting PII from unstructured text using AI_REDACT, released in General Availability in December 2025.

AI_REDACT is a fully-managed Cortex AI Function that uses a large language model (LLM) to detect, locate, and redact PII from unstructured text data. The function is available natively in select regions (AWS US, AWS Europe, Azure US/Europe). If unavailable in your region, you must enable cross-region inference. Organizations with strict data residency requirements should verify regional availability or evaluate whether cross-region inference aligns with compliance policies before deployment.

• Important things to know: AI_REDACT and Dynamic Data Masking serve complementary purposes. Dynamic Data Masking protects structured PII columns (SSN, EMAIL, phone numbers). AI_REDACT targets unstructured text fields where PII appears alongside other content, feedback, notes, or messages, where column-level masking is not applicable.

The Scenario

Your organization collects customer feedback, support tickets, and survey responses containing valuable insights, but also personally identifiable information (PII) like names, emails, phone numbers, and addresses. You need to analyze this data for trends and sentiment, but GDPR, HIPAA, and other regulations require that PII be removed first.

Traditional approach: Build complex regex patterns for each PII type, manually review edge cases, and constantly update rules as data formats evolve or use expensive third-party tools.

With Cortex AI_REDACT: Automatically detect and redact PII using AI in a single SQL function, with no regex patterns or external services required.

Step 1: Prepare Your Data

Create a table with customer feedback containing various types of PII embedded in free-form text. This simulates real-world data where customers naturally include personal information when providing feedback.

#sql

-- Customer feedback with PII embedded in free-form text

CREATE OR REPLACE TABLE raw_customer_feedback (

feedback_id INTEGER

,feedback_text TEXT

,created_at TIMESTAMP_NTZ

);

-- Insert sample feedback with various types of PII

INSERT INTO raw_customer_feedback VALUES

(1, 'Hi, my name is Sarah Johnson and I love your product! You can reach me at sarah.j@email.com or call 415-555-0123 if you have questions.', '2026-01-20 10:15:00')

,(2, 'John Smith here, living at 123 Main Street, San Francisco, CA 94102. The delivery was late but the product quality is excellent. My order number was ORD-98234.', '2026-01-20 11:30:00')

,(3, 'Terrible experience! I called support and gave them my SSN 123-45-6789 and credit card 4532-1234-5678-9010. Still no resolution after 3 weeks.', '2026-01-20 14:45:00')

,(4, 'Amazing service! Emma Wilson from Boston here. The team helped me within minutes. Highly recommend!', '2026-01-20 16:20:00')

,(5, 'Product arrived damaged. I am Michael Chen, customer ID CUST-55421. Please contact me at (650) 555-7890 or mchen@company.com to arrange replacement.', '2026-01-21 09:00:00')

;

Step 2: Redact All PII with AI_REDACT

Use AI_REDACT to automatically identify and mask all types of PII in the text. The AI understands context; it knows "Sarah Johnson" is a name, "415-555-0123" is a phone number, and "123 Main Street, San Francisco, CA 94102" is an address, even written in natural language.

#SQL

CREATE OR REPLACE TABLE redact_customer_feedback

as

SELECT

feedback_id,

SNOWFLAKE.CORTEX.AI_REDACT(feedback_text) AS redacted_text,

created_at

FROM raw_customer_feedback;

For some use cases, you may only need to redact specific PII categories. AI_REDACT allows you to specify which types to mask, leaving other information visible. This is useful when you need to preserve certain data for analysis while protecting the most sensitive information.

#SQL

SELECT

feedback_id,

SNOWFLAKE.CORTEX.AI_REDACT(

feedback_text,

ARRAY_CONSTRUCT('NAME', 'EMAIL', 'PHONE_NUMBER')

) AS redacted_text,

created_at

FROM raw_customer_feedback;

Step 3: Privacy-Safe Sentiment Analysis

Combine AI_REDACT with other Cortex AI functions to perform analytics while maintaining privacy compliance. Here, we analyze sentiment on redacted feedback, ensuring PII never enters our analytics pipeline.

#SQL

SELECT

feedback_id,

redacted_text,

SNOWFLAKE.CORTEX.AI_SENTIMENT(

redacted_text

) AS sentiment_score

FROM redact_customer_feedback;

Considerations When Using AI_REDACT

While AI_REDACT significantly simplifies PII detection, consider these points: The function currently works best with well-formed English text. AI-based detection may occasionally produce false positives (redacting non-PII) or false negatives (missing PII), so validate results on your specific data patterns. For mission-critical compliance, combine AI_REDACT with traditional validation methods during initial implementation. Always test thoroughly with representative samples before production deployment to ensure compliance with your organization’s policies.

AI_REDACT Performance and Cost Optimization

AI_REDACT is optimized for throughput and batch processing. Snowflake recommends processing numerous inputs from large SQL tables rather than row-by-row operations. Batch processing is typically better suited for AI Functions. Use a warehouse size no larger than MEDIUM when calling AI_REDACT, as larger warehouses do not increase performance but can result in unnecessary costs.

• Result: you've built a privacy-compliant analytics pipeline that automatically strips PII before sentiment analysis, no regex patterns, no manual review, and no external services required. The AI understands context and handles variations like "my name is X", "X here", addresses written naturally, and more.

Key Advantages of Cortex AI

No data movement: Process sensitive data in-place, maintaining compliance and governance without exfiltration.

Enterprise security: Inherits Snowflake's RBAC, data masking, and audit capabilities. Models don't train on your data.

Model flexibility: Access multiple LLMs (Claude, GPT-4, Mistral, Llama, Gemini) through a unified interface.

Cost efficiency: Pay-per-use credits. No separate infrastructure or minimum commitments.

Seamless integration: Works with existing Snowflake workflows and BI platforms.

Considerations & Limitations of Snowflake Cortex AI

Platform dependency: Requires data to reside in Snowflake to fully leverage capabilities.

Limited customization: Less control over model tuning compared to dedicated ML platforms.

LLM variability: Outputs may be non-deterministic and require validation for critical use cases.

Data readiness required: Effectiveness depends on well-structured, governed data.

Cost variability: Credit-based pricing can scale quickly without proper monitoring.

Regional availability: Not all functions and models are available in every region; some require cross-region inference with data residency implications.

Feature maturity: Some capabilities are still evolving and may lack enterprise depth. Verify GA status before production deployment.

Pricing and Cost Management

Cortex AI uses credit-based consumption pricing, varying by model, operation, and tools used. It is important to monitor regularly the cost and disable the service unused.

There are a variety of things that drive cost in Cortex AI, including:

Model selection: Larger models (Claude, GPT-4) cost significantly more than smaller models (Mistral 7B, Gemma)

Cortex Search: Charges accrue continuously even without queries, always suspend or drop unused services

Data volume: Processing more text increases costs proportionally

Monitor consumption: Monitor usage regularly to identify and eliminate waste.

Tracking cost at organization-levels

SELECT SERVICE_TYPE, SUM(credits_used) AS total_credits

FROM SNOWFLAKE.ORGANIZATION_USAGE.METERING_DAILY_HISTORY

WHERE service_type ilike '%AI_SERVICES%'

AND usage_date >= DATEADD(day, -30, CURRENT_DATE()) --Last Month

GROUP BY SERVICE_TYPE;

For a breakdown of costs by specific AI service, refer to the following account usage views:

• SNOWFLAKE.ACCOUNT_USAGE.CORTEX_AISQL_USAGE_HISTORY

• SNOWFLAKE.ACCOUNT_USAGE.CORTEX_ANALYST_USAGE_HISTORY

• SNOWFLAKE.ACCOUNT_USAGE.DOCUMENT_AI_USAGE_HISTORY

• SNOWFLAKE.ACCOUNT_USAGE.CORTEX_SEARCH_SERVING_USAGE_HISTORY

Conclusion

Snowflake Cortex AI makes enterprise-grade artificial intelligence accessible and practical for data teams. By bringing LLMs, search, and analytics directly into the platform, it eliminates integration complexity while respecting governance requirements. Cortex AI works seamlessly with both structured data (tables, SQL) and unstructured data (text, documents), enabling comprehensive AI-powered insights across your entire data estate.

Our AI_REDACT example demonstrated how easily you can build privacy-compliant analytics pipelines. The function uses AI to understand context and identify PII in natural language, handling variations like "my name is X", addresses written naturally, and more without brittle regex patterns. Combine AI_REDACT with other Cortex functions like AI_SENTIMENT, AI_COMPLETE, or AI_SUMMARIZE to safely analyze sensitive text while maintaining GDPR, HIPAA, and CCPA compliance.

Keyrus & Snowflake

As a Premier Snowflake Partner, we help organizations adopt Cortex Code and Cortex AI in a coordinated manner, ensuring that productivity gains do not outpace governance maturity. Our approach aligns platform engineering, FinOps monitoring, and AI enablement to create sustainable enterprise AI architectures within Snowflake. Learn more about our partnership or contact us to start your Coretex ecosystem.