<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Angela Harney on Medium]]></title>
        <description><![CDATA[Stories by Angela Harney on Medium]]></description>
        <link>https://medium.com/@angelamarieharney?source=rss-7e068c9737ea------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*j59knOklM9h26VjrmPU6Qg.jpeg</url>
            <title>Stories by Angela Harney on Medium</title>
            <link>https://medium.com/@angelamarieharney?source=rss-7e068c9737ea------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 16:18:54 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@angelamarieharney/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Cortex Code Skills — First Steps]]></title>
            <link>https://medium.com/@angelamarieharney/cortex-code-skills-first-steps-1a4cc7e5d88f?source=rss-7e068c9737ea------2</link>
            <guid isPermaLink="false">https://medium.com/p/1a4cc7e5d88f</guid>
            <dc:creator><![CDATA[Angela Harney]]></dc:creator>
            <pubDate>Mon, 20 Apr 2026 02:34:59 GMT</pubDate>
            <atom:updated>2026-04-20T03:45:02.060Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/538/1*_i4xul8Xpyyb9uGX3RBwaA.jpeg" /></figure><h3>How to Create Skills for Cortex Code in Snowsight</h3><p>A beginner-friendly tutorial in 8 steps.</p><h3>Step 1: What Is a Skill?</h3><p>A <strong>skill</strong> is a small instruction file that teaches the AI assistant in Snowsight Workspaces called Cortex Code (CoCo) how to perform a specific task. Instead of explaining what you want every time, you write the instructions once, and the AI follows them whenever you (or anyone) asks.</p><p>A skill is just a text file called SKILL.md placed in a special folder in your workspace. It can be used to distribute workflows to your team.</p><p><strong>Key concepts:</strong></p><ul><li>A skill does NOT run code by itself — it tells CoCo <em>what to do</em> and <em>how to do it</em></li><li>Skills can reference stored procedures, SQL patterns, or any repeatable workflow</li><li>Anyone with the SKILL.md file can install it in their own workspace</li></ul><h3>Step 2: Markdown Basics</h3><p>SKILL.md files are written in <strong>Markdown</strong>, a simple formatting language. Here is everything you need to know:</p><h3>Headings</h3><p>Use # symbols to create headings. More # signs = smaller subheading.</p><pre># Heading 1 (biggest)<br>## Heading 2 (section)<br>### Heading 3 (subsection)<br>#### Heading 4 (detail)<br>##### Heading 5<br>###### Heading 6 (smallest)</pre><p>For skills, stick to # and ##. Use ### sparingly.</p><h3>Bold and Italic</h3><pre>**bold text**<br>*italic text*</pre><h3>Bullet Lists</h3><pre>- Item one<br>- Item two<br>- Item three</pre><h3>Numbered Lists</h3><pre>1. First step<br>2. Second step<br>3. Third step</pre><h3>Code (Inline and Blocks)</h3><p>Inline code uses single backticks. Note: a stored procedure is optional.</p><pre>Call the `my_procedure` stored procedure.</pre><p>Code blocks use triple backticks with an optional language tag:</p><pre>```sql<br>SELECT * FROM my_table;<br>```</pre><h3>Links</h3><pre>[Link text](https://example.com)</pre><h3>Horizontal Rule</h3><pre>---</pre><p>That’s it. You now know enough Markdown to write any skill file.</p><h3>Step 3: The SKILL.md File Structure</h3><p>Every SKILL.md has two parts:</p><h3>Part 1: Front Matter (the header between --- lines)</h3><pre>---<br>name: my-skill-name<br>description: A short sentence explaining when this skill should be used.<br>---</pre><ul><li><strong>name</strong>: A unique identifier (use lowercase and hyphens, no spaces)</li><li><strong>description</strong>: This is what Cortex Code reads to decide if your skill matches the user’s request. Make it specific.</li></ul><h3>Part 2: Body (the instructions)</h3><p>The body tells Cortex Code what to do. A good structure is:</p><pre># When to Use<br>- Scenario one<br>- Scenario two</pre><pre># Instructions<br>1. Step one<br>2. Step two</pre><pre>## Parameters<br>- `param_1`: Description of the parameter</pre><pre># Examples<br>## Example 1: Short title<br>User: What the user might say</pre><pre>## Assistant<br>The response Cortex Code should give (often a SQL block)</pre><p><strong>Important:</strong> The # Examples section with User: and ## Assistant pairs teaches Cortex Code by example. Always include at least one.</p><h3>Step 4: Where to Put the File</h3><p>Skills must be placed in a specific folder path inside your Snowsight Workspace:</p><pre>.snowflake/cortex/skills/&lt;your-skill-name&gt;/SKILL.md</pre><h3>Step-by-step (manual, no AI credits):</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/889/1*2R1lp82kVcx6HXN7i6KP4g.png" /></figure><h3>Alternative (using Cortex Code):</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*7rWECcgKZMj3nlUoBrHpPA.png" /></figure><h3>Step 5: Write Your First Skill</h3><p>Let’s build a simple skill from scratch. This one will help users quickly generate a row count summary for any table.</p><p>Create the file .snowflake/cortex/skills/row-count-check/SKILL.md with this content:</p><pre>---<br>name: row-count-check<br>description: Generates a SQL query to check the row count of a specified table.<br>---</pre><pre># When to Use<br>- When a user wants to quickly check how many rows are in a table<br>- When validating that data loaded correctly</pre><pre># Instructions<br>1. Ask the user which table they want to check (fully qualified: DATABASE.SCHEMA.TABLE)<br>2. Run a SELECT COUNT(*) query against that table<br>3. Report the result</pre><pre># Examples<br>## Example 1: Check a table<br>User: Use the row-count-check skill on ANALYTICS.PUBLIC.SALES</pre><pre>## Assistant<br>```sql<br>SELECT COUNT(*) AS ROW_COUNT FROM ANALYTICS.PUBLIC.SALES;<br>```</pre><p>Save the file. Now when you open a new Cortex Code session in that workspace, you can say:</p><blockquote><em>“Use the row-count-check skill on MY_DB.MY_SCHEMA.MY_TABLE”</em></blockquote><p>And it will follow your instructions.</p><h3>Step 6: Using a Skill</h3><p>Once your skill is installed, here is how to actually use it in Cortex Code.</p><h3>Starting a session</h3><ol><li>Open your <strong>Workspace</strong> in Snowsight</li><li>Click the <strong>Cortex Code</strong> icon (the AI assistant panel) to start a new session</li></ol><h3>Invoking the skill</h3><p>There are two ways to trigger a skill:</p><p><strong>Option A: Ask by name</strong></p><p>Type a message that references the skill name directly:</p><blockquote><em>“Use the row-count-check skill on ANALYTICS.PUBLIC.SALES”</em></blockquote><p>Cortex Code will match the name, load the SKILL.md, and follow the instructions inside it.</p><p><strong>Option B: Describe what you need</strong></p><p>You can also just describe the task. If your skill’s description field in the front matter is specific enough, Cortex Code will automatically match and load it:</p><blockquote><em>“How many rows are in ANALYTICS.PUBLIC.SALES?”</em></blockquote><p>This works because the skill’s description mentions checking row counts.</p><h3>What happens when a skill runs</h3><ol><li>Cortex Code reads the entire SKILL.md content</li><li>It follows the # Instructions section step by step</li><li>It uses the # Examples section as a pattern for its response</li><li>If the skill references a stored procedure, Cortex Code will generate and execute the CALL statement</li></ol><h3>Verifying the skill loaded</h3><p>When a skill is invoked, you will see a message in the chat like:</p><blockquote><em>The “row-count-check” skill is running</em></blockquote><p>If you do not see this, the skill was not matched. Check that:</p><ul><li>The file is named exactly SKILL.md</li><li>It is in the correct path: .snowflake/cortex/skills/&lt;name&gt;/SKILL.md</li><li>The description in the front matter is relevant to what you asked</li><li>You started a <strong>new</strong> CoCo session after adding the file</li></ul><h3>Tips</h3><ul><li><em>You can have multiple skills installed at once — they do not conflict</em></li><li><em>If a skill is not triggering automatically, use the explicit name: “Use the &lt;skill-name&gt; skill”</em></li><li><em>Skills persist across sessions as long as the file stays in the workspace</em></li><li><em>To see a list of installed skills type /skills in the CoCo prompt.</em></li></ul><h3>Step 7: Distributing Skills to Others</h3><p>When you want to share a skill with someone else, here is what to give them:</p><h3>What to include:</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*DIcppdG6boGk6fidaz3mmg.png" /></figure><h3>Folder structure to share:</h3><pre>my-skill/<br>├── SKILL.md                 # Required - Cortex Code reads this<br>├── README.md                # Optional - setup guide for humans<br>└── scripts/<br>    └── setup.sql            # Optional - stored procedure / DDL</pre><h3>Installation instructions to give recipients:</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MVvxCawcOw-nS0Cs6-V6tA.png" /></figure><h3>Important notes:</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*v8q4W0iYNHw9VbCznnubLw.png" /></figure><h3>Step 8 (Optional): How to add Human-in-the-Loop in Skills that take actions</h3><p>When your skill calls a stored procedure or takes an action, the AI will execute it immediately. Stored procedures can create objects, modify data, or consume credits — so you should build <strong>stop markers</strong> into your skill to pause and confirm before anything runs.</p><h3>What is a stop marker?</h3><p>A stop marker is an instruction in your SKILL.md that tells Cortex Code to <strong>stop and ask the user for confirmation</strong> before executing a destructive or expensive action. Without one, the AI may run the procedure the moment the skill is triggered.</p><h3>How to add stop markers</h3><p>Add a confirmation step in your # Instructions section. Here is the pattern:</p><pre># Instructions<br>1. Gather the required parameters from the user<br>2. Show the user the exact CALL statement that will be executed<br>3. **STOP. Ask the user to confirm before running.** Do NOT execute until the user says yes.<br>4. Once confirmed, execute the CALL statement</pre><p>The key is step 3. Bold text and explicit language like “STOP” and “Do NOT execute” make it clear to the AI.</p><h3>Full example with a stop marker</h3><pre>---<br>name: enrich-table<br>description: Enriches a table with AI sentiment and classification columns using a stored procedure.<br>---</pre><pre># When to Use<br>- When a user wants to add AI-generated columns to a raw data table</pre><pre># Instructions<br>1. Ask the user for the source table name and target table name<br>2. Ask which text columns should be analyzed for sentiment<br>3. Ask what classification labels to apply<br>4. Generate the CALL statement with the provided parameters<br>5. **STOP. Display the CALL statement and ask the user to confirm before executing.**<br>   Do NOT run the procedure until the user explicitly approves.<br>6. Once the user confirms, execute the CALL statement<br>7. Report the results</pre><pre># Examples<br>## Example 1: Enrich with confirmation<br>User: Use the enrich-table skill on RAW_FEEDBACK</pre><pre>## Assistant<br>Here is the CALL statement I will execute:</pre><pre>```sql<br>CALL create_advanced_ai_enrichment_skill(<br>    &#39;RAW_FEEDBACK&#39;,<br>    &#39;ENRICHED_FEEDBACK&#39;,<br>    ARRAY_CONSTRUCT(&#39;COMMENT&#39;),<br>    ARRAY_CONSTRUCT(&#39;BILLING&#39;, &#39;SUPPORT&#39;),<br>    ARRAY_CONSTRUCT(<br>        OBJECT_CONSTRUCT(&#39;column&#39;, &#39;COMMENT&#39;, &#39;labels&#39;, ARRAY_CONSTRUCT(&#39;BUG_REPORT&#39;, &#39;FEATURE_REQUEST&#39;))<br>    )<br>);<br>```</pre><pre>Shall I run this?</pre><h3>When to use stop markers</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1kJYnQzD0EvsdEAHK6aHHw.png" /></figure><h3>Additional safety patterns</h3><p>You can also add these instructions to your skill:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*odxXkFESG82qdk38bs4HTg.png" /></figure><p>These patterns give users control and visibility before anything permanent happens.</p><h3>Quick Reference Notes</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*s0N3MOGAo7Mx4W3D-nFVrQ.png" /></figure><p>It’s easy and fun! Come and join the crowd!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1a4cc7e5d88f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Strategy document to streamline adoption using Cortex AI]]></title>
            <link>https://medium.com/@angelamarieharney/ai-strategy-document-that-uses-cortex-ai-to-streamline-adoption-bf768a0e3f06?source=rss-7e068c9737ea------2</link>
            <guid isPermaLink="false">https://medium.com/p/bf768a0e3f06</guid>
            <dc:creator><![CDATA[Angela Harney]]></dc:creator>
            <pubDate>Tue, 03 Mar 2026 07:39:41 GMT</pubDate>
            <atom:updated>2026-06-15T19:44:13.056Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Work Smarter not Harder at Every Level</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/605/1*yCOzjefluOphfqQSWfJCLg.png" /></figure><p>Index<br>- How to Overcome Obstacles<br>- AI Philosophy<br>- AI Blueprint<br>- AI Strategy Document Overview<br>- AI Powered Analytics<br>- Next Steps</p><h3>Overcoming Obstacles</h3><p>All companies face the same challenges when adopting new programs:</p><blockquote>1. Executive buy-in</blockquote><blockquote>2. Employee adoption</blockquote><blockquote>3. Adding new processes to already over-burdened departmental teams that require measuring performance, training of new materials and feedback to improve accuracy</blockquote><p>Adopting AI insights into actionable analytics sounds good, but quickly becomes too small of an ROI for the amount of change it requires within an organization… <strong>UNLESS ….</strong></p><ul><li><em>Executives hear that 3 business decisions are chosen as data products and are broken down into actionable components</em></li><li><em>Managers hear that they don’t have to receive additional alerts, review more reports and dashboards</em></li><li><em>Employees hear that AI won’t replace their job, it’s here to enable them to Work Smarter not Harder</em></li><li><em>Teams hear that they only have to interact with a single Streamlit application interface showing a project status map and a prompt where they can ask relative AI insights for their role</em></li></ul><h3>AI Philosophy</h3><p><strong>Work Smarter not Harder at Every Level</strong> is an AI Philosophy that brings a single prompt interface to every level of engagement to address ease of adoption with a strategy of focusing on answering business decisions.</p><p>A Streamlit app has been designed with a color-coded set of gears as an image representing different aspects of project status next to a single prompt where...</p><ul><li><em>An executive can ask what level of decisions the system is able to provide as the strategy matures, as well as if there are cost overruns.</em></li><li><em>A manager can ask progress on delivery of AI data products, as well as what employees are lacking in AI learning, adoption and enablement.</em></li><li><em>An employee can ask what AI learning they need to catch up on, as well as how to apply the newest insights to a task they are about to do so they can provide qualitative input about efficiencies gained.</em></li></ul><p>The AI Intelligence prompt / scorecard Streamlit application that leverages Cortex AI for insights gets rid of the need for dashboards, reporting alerts, utilization reports and even some meetings.</p><h3>AI Blueprint</h3><p>Each business decision is a delivered as a data product that is broken down into components, making them practical and reasonable to accomplish.</p><p>The first business decision is used to establish the groundwork. Each component consists of:</p><ol><li>Identify Business Decisions mapped to KPIs</li><li>Identify Project Tracking KPIs</li><li>Identify Learning</li><li>Identify Enablement</li><li>Identify AI Cost Measurement</li></ol><p>Enablement is provided in micro-learning sessions that are short and targeted.</p><p>Components can be larger or small and delivered in parallel with existing work streams based on capacity.</p><h3>AI Strategy Document Overview</h3><p>The strategy is outlined below in a model that can be tailored to any project. The AI strategy document itself is designed so that each section is an actionable reference to illustrate the step to related teams.</p><p>There is a lot of information condensed into a practical and actionable format that can be delivered in reasonable timeframes.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/871/1*qz5kmfGnB9WftADqtsq4Bw.png" /></figure><h3>AI Powered Analytics</h3><p>If the application to adopt AI insights and ensure it stays on track is this easy, then anyone at Every Level can do it.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*9JuPEgYyqbCDJP2wcaJwvA.png" /></figure><h3>Next Steps</h3><p>Align your organization to adopt AI insights internally and be positioned to weather the economic challenges ahead.</p><p>Contact me on LinkedIn to ask about my AI Strategy Document that uses Cortex AI to streamline adoption.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bf768a0e3f06" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Cortex AI Cost Queries in Snowflake]]></title>
            <link>https://medium.com/@angelamarieharney/cortex-ai-cost-queries-in-snowflake-07331811d42d?source=rss-7e068c9737ea------2</link>
            <guid isPermaLink="false">https://medium.com/p/07331811d42d</guid>
            <dc:creator><![CDATA[Angela Harney]]></dc:creator>
            <pubDate>Mon, 19 Jan 2026 05:21:47 GMT</pubDate>
            <atom:updated>2026-04-22T22:04:15.830Z</atom:updated>
            <content:encoded><![CDATA[<h4>Know how to track your AI Costs and increase Trust</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/679/1*6s6nS3MNTNo0aJ_tdd0fiQ.png" /></figure><blockquote><strong>Contents</strong><br> —<a href="https://medium.com/p/07331811d42d#f490"> Ingest Costs</a><br> — <a href="https://medium.com/p/07331811d42d#9d66">Ingress Costs</a><br> — <a href="https://medium.com/p/07331811d42d#143d">Egress Costs</a><br> —<a href="https://medium.com/p/07331811d42d#b99c"> Compute Costs</a><br> — <a href="https://medium.com/p/07331811d42d#e547">Counting Tokens</a><br> — <a href="https://medium.com/p/07331811d42d#5e4a">Monthly Invoice Amount</a><br> — <a href="https://medium.com/p/07331811d42d#714b">Cortex AI Usage History Views</a><br> — <a href="https://medium.com/p/07331811d42d#870d">My Cost Queries</a><br> — <a href="https://medium.com/p/07331811d42d#7391">Other Cost Considerations and Monitoring</a><br> — <a href="https://medium.com/p/07331811d42d#6316">Cost Summary</a><br> — <a href="https://medium.com/p/07331811d42d#d3d0">Cortex Code Costs</a> <br>— <a href="https://medium.com/p/07331811d42d#fc36">Cortex Code UI Costs</a><br> — <a href="https://medium.com/p/07331811d42d#fc36">Cortex Code CLI Costs</a><br> — <a href="https://medium.com/p/07331811d42d#3dea">Cortex Search Costs</a><br> — <a href="https://medium.com/p/07331811d42d#2326">Appendix I: Invoice Cost Details</a><br> — <a href="https://medium.com/p/07331811d42d#ae3c">Appendix II: Individual AI Query Details</a><br> — <a href="https://medium.com/p/07331811d42d#bfbc">Appendix III: Second-Level Cost Details</a> <br>— <a href="https://medium.com/p/07331811d42d#df73">Appendix IV: Cortex Code Cost Details</a></blockquote><h3>Introduction</h3><p>I recently had to provide a breakdown of the AI services costs that are listed on the monthly Snowflake invoice so that it was clearly understood where AI costs were being attributed to help make planning decisions.</p><p>Snowflake’s integration of AI services like Cortex AI introduces new dimensions to cost management and understanding AI costs.</p><blockquote>AI costs fall into four main categories: ingest, inference, standard compute, and egress.</blockquote><h3>Ingest Costs</h3><p>Ingest costs arise when data is processed by AI models, such as during document parsing or text analysis via Cortex Document AI.</p><p>These costs are tracked in the CORTEX_DOCUMENT_PROCESSING_USAGE_HISTORY view, where CREDITS_USED reflects the compute consumed per operation.</p><h3>Inference Costs</h3><p>Inference costs occur when AI models are invoked to generate insights, such as through Cortex Search or AI-driven query suggestions or when calling functions like AI_COMPLETE or AI_EMBED and are billed per token.</p><p>These can be monitored using the CORTEX_INFERENCE_USAGE_HISTORY view. To control costs, teams should add query tags to AI queries, limit high-volume operations, and use serverless compute wisely, especially for one-off AI tasks.</p><p>Generative functions charge for input and output tokens, whereas Embedding and Similarity functions only charge for input tokens.</p><p>These costs are tracked in service-specific usage views and are rolled up into the AI_INFERENCE line item on the Snowflake monthly invoice.</p><h3>Compute Costs</h3><p>Standard compute costs for AI workloads are tracked separately in WAREHOUSE_METERING_HISTORY.</p><p>While AI processing uses serverless AI Services compute, the warehouse used to orchestrate the query, such as when usingAI_COMPLETE, incurs standard compute charges.</p><p>This cost is independent of AI token usage and must be monitored alongside AI Services credits.</p><h3>Egress Costs</h3><p>Egress costs are a critical but often underestimated component of AI workloads. Monitoring CLOUD_SERVICES_USAGE_HISTORY can help identify high egress patterns.</p><p>While Snowflake offers zero egress cost for cross-region and cross-cloud data sharing via the Egress Cost Optimizer (ECO) which eliminates cross-region and cross-cloud data transfer fees for shared data helps minimize egress costs. Manual replication or data exported to external systems using COPY INTOcan incur significant fees.</p><p>AI models that generate large outputs (e.g., summaries, embeddings) and transfer them outside Snowflake will trigger egress charges.</p><p>To mitigate this, use internal data sharing, limit output size, and ensure AI outputs are consumed within the same cloud region.</p><h3>Counting Tokens</h3><blockquote>The <em>AI_COUNT_TOKENS</em> function plays a critical role in cost estimation and optimization.</blockquote><p>It calculates the number of input tokens a given text will consume when processed by a specific AI model, allowing teams to estimate costs before execution, similar to using Explain on a SQL query.</p><p>This is especially valuable for prompt engineering and avoiding model limits.</p><p>No AI token charges apply when usingAI_COUNT_TOKENS.It only incurs standard compute cost, so it is great to use in pre-validating high-volume AI workflows.</p><p>In this example, it is used in a select query over a table that stores document chunks in columns. Putting document chunks in separate columns takes advantage of Snowflake’s columnar optimization and limits the amount of data scanned for the AI Function being used. This approach is setup to run data through a standard data pipeline.</p><pre>select<br>    SNOWFLAKE.CORTEX.COUNT_TOKENS(&#39;llama3-8b&#39;,doc_contents) AS token_count, token_count * (.0003/1000) as token_cost;<br>    count_tokens(<br>          , ai_complete<br>          , &#39;llama3-8b&#39;,<br>          , &lt;&lt; column_name that stores the document or chunk to assess &gt;&gt;<br>          , &#39;What are the revenue projections for the next 5 years based on the last two years profit.&#39;<br>        )<br>FROM my_table_name<br>;</pre><h3>Monthly Invoice Amount</h3><p>To get the view of AI costs that will be reflected on your Snowflake invoice at the end of the month, throughout the month you can query the METERING_DAILY_HISTORY view filtered by service type.</p><p>This view contains an aggregation of all AI-related credit consumption and aligns with billing statements. Multiply CREDITS_USED by your negotiated credit rate to estimate actual spend.</p><pre>SELECT <br>  USAGE_DATE,<br>  SERVICE_TYPE,<br>  CREDITS_USED,<br>  CREDITS_USED * &lt;your_credit_rate&gt; AS estimated_cost_usd<br>FROM SNOWFLAKE.ORGANIZATION_USAGE.METERING_DAILY_HISTORY<br>WHERE SERVICE_TYPE IN (&#39;AI_SERVICES&#39;, &#39;AI_INFERENCE&#39;)<br>  AND USAGE_DATE &gt;= DATEADD(&#39;month&#39;, -3, CURRENT_DATE())<br>ORDER BY USAGE_DATE DESC;</pre><h3>Cortex Usage History Views</h3><p>Cortex AI costs are tracked across different levels in dedicated ACCOUNT_USAGE views, each providing granular insights into credit consumption. These views enable monitoring at multiple levels, from high-level trends to individual query costs. Understanding the basics of how usage costs are grouped will help you execute cost queries to review the breakouts.</p><p><em>This is not a detailed explanation of all Cortex Usage History views. See cost queries in the Appendixes for alphabetical list of Cortex Usage History views broken out by first-level and second-level views.</em></p><h4>AISQL <strong>Functions</strong></h4><p><em>The </em>CORTEX_FUNCTIONS_QUERY_USAGE_HISTORY<em>view provides individual query-level details, including </em><em>query_id, </em><em>user_name, </em><em>warehouse_name, </em><em>function_name, </em><em>model_name, </em><em>tokens, and </em><em>token_credits. This view is essential for pinpointing expensive queries and optimizing performance and is a first-level consolidation cost on invoices.</em></p><h4>Cortex Analyst</h4><p><em>Usage is metered at a secondary level in the </em>CORTEX_ANALYST_USAGE_HISTORY<em>view and is aggregated hourly, showing credits consumed and number of messages per user. It does not include </em><em>query_id, but offers per-user visibility into AI interaction costs. Costs are incurred per successful response (message) not by token count.</em></p><h4>Cortex Search</h4><p><em>Usage is metered at a secondary level in the </em>CORTEX_SEARCH_DAILY_USAGE_HISTORY <em>view and breaks down costs daily by consumption type (</em><em>serving, </em><em>embed_text_tokens) and includes costs for storage, embedding, and serving compute as follows:</em></p><ul><li>CORTEX_SEARCH_DAILY_USAGE_HISTORY (serving vs. embedding)</li><li>CORTEX_SEARCH_SERVING_USAGE_HISTORY (hourly serving credits)</li></ul><h4>Metering History</h4><p><em>A broader view that can be filtered by </em><em>SERVICE_TYPE = &#39;AI_SERVICES&#39; to track overall AI credit usage across all Cortex services, including AI functions, Analyst, and Search.</em></p><h3>My Cost Queries</h3><p>My Snowflake cost queries are broken out by the following practical uses and are filtered to the beginning of the month for 60 days ago including the current month-to-date. So, mid-way through January, the costs for Nov / Dec / Jan will be returned.</p><blockquote><strong>Invoice Amounts</strong></blockquote><ul><li><strong>Invoice Summary</strong>: <br><em>Invoice line-item alignment</em></li><li><strong>Invoice (First-Level) Detail</strong>: <br><em>First-level Cortex usage history views that are consolidated within the invoice line items, and also include a standard Compute aggregation from virtual warehouse costs</em></li></ul><blockquote><strong>Activity Details</strong></blockquote><ul><li><strong>Individual AI Query Detail</strong>: <br><em>Cost and performance stats for individual Cortex AI queries</em></li><li><strong>Other Tracking (Second-Level) Detail</strong>: <br><em>Second-level Cortex usage history views not consolidated for invoicing, but are used for tracking specific activities</em></li></ul><h4>Invoice Summary</h4><p>Use the following query to isolate warehouse costs at an Organizational billing level to tie-out to the invoice cost for AI INFERENCE and AI SERVICES that are shown in the Snowflake Monthly Invoice Statement.</p><pre>SELECT<br>      TO_DATE(USAGE_DATE) AS USAGE_DATE,<br>      &#39;INVOICE SUMMARY&#39; AS SERVICE_GROUP,<br>      SERVICE_TYPE,<br>      SUM(CREDITS_USED) as CREDITS_USED,<br>      SUM(CREDITS_USED) * &lt;&lt; your credit price &gt;&gt; AS INVOICE_COST,<br>      0 as TOKENS<br>FROM snowflake.organization_usage.metering_daily_history h<br>WHERE service_type IN (&#39;AI_SERVICES&#39;,&#39;AI_INFERENCE&#39;) <br>AND TO_DATE(USAGE_DATE) &gt;= DATE_TRUNC(&#39;MONTH&#39;,DATEADD(day, -60, CURRENT_DATE))<br>GROUP BY ALL<br>ORDER BY 1<br>;</pre><h4>Invoice and First-Level Details</h4><p>In my invoice detail query, AI Inference breaks out to document processing, AI Services breaks out to function usage and there is an aggregation section for Compute as standard compute for virtual warehouse costs. See APPENDIX I — Invoice Cost Details.</p><p><em>Note: Additional Cortex Usage History views may be added as first-level charges that aggregate to invoice line-items over time as Snowflake enhances their offerings.</em></p><h4>Individual AI Query Details</h4><p>In addition to invoice costs, I also wanted to know how individual AI queries performed, and if they scanned more than the amount of data that I expected.</p><p>I created the query listed in APPENDIX II — Individual AI Query Details to return a user-friendly selection and display of columns that focus on assessing individual query executions, along with basic execution stats, and includes columns such as invocations and row counts for assessing how much data was actually scanned.</p><p>The table listed in the JOIN clause can be swapped out for any of the Cortex usage history views that contain a query ID.</p><h4>Second-Level Details</h4><p>Second-level charges provided by Snowflake for tracking of specific activities can be found in my query in Appendix II-APPENDIX III — Second-Level Cost Details.</p><p>For example, although the cortex_search_history_usage view is consolidated to an invoice line item, the secondary-level cortex_search_serving_usage_historyview is next-level details for Cortex Search activities.</p><p><em>Note: Additional second-level Cortex Usage History views may be added over time as Snowflake enhances their offerings.</em></p><h3>Other Cost Considerations and Monitoring</h3><p>Querying AI costs using Snowflake’s native monitoring tools covers several features and areas of functionality.</p><h4>Notebooks</h4><p>The ACCOUNT_USAGE schema provides granular visibility into AI-related credit consumption. For example, running a query against SNOWFLAKE.ACCOUNT_USAGE.CORTEX_DOCUMENT_PROCESSING_USAGE_HISTORY allows you to identify which Notebooks or users trigger the highest AI processing costs.</p><p>Similarly, NOTEBOOKS_CONTAINER_RUNTIME_HISTORY helps track AI workload duration and credit usage for Snowflake Notebooks with AI components.</p><p>By combining these views with object and query tags, teams can attribute AI costs to specific projects, departments, or users that enable accurate budgeting and chargeback models.</p><h4>Containers</h4><p>Containers in Snowflake, powered by Snowpark Container Services (SPCS), directly impact AI costs by providing dedicated compute pools (ranging from CPU to GPU), that can be selected for running containerized AI/ML workloads such as fine-tuning LLMs, distributed embedding generation, or deploying custom AI models.</p><p>Unlike standard virtual warehouses, container runtime costs are based on the uptime of the compute pool (small, medium, or GPU-enabled) and are billed per credit hour, with pricing varying by instance type and region.</p><p>Running AI notebooks or Streamlit apps using container runtime (instead of warehouse runtime) can be more cost-effective, especially for GPU-intensive tasks, since Snowflake manages the underlying infrastructure with no additional fees for registry, networking, or logs.</p><p>However, long-running or oversized containers can quickly increase costs, so it’s essential to monitor CONTAINER_SERVICES_MONITORING_HISTORY and use auto-suspend settings.</p><h4>Query Patterns and Resource Allocation</h4><p>Effective cost control still hinges on optimizing query patterns and resource allocation like all other data querying does.</p><p>A common mistake is relying on simple duration-based cost estimates, which ignore concurrency and auto-suspend delays. Instead, use a normalized, warehouse-uptime-based method to allocate costs. Treat each warehouse runtime as a single event, then distribute costs proportionally based on query duration within that window.</p><p>This method accounts for idle time and concurrency, revealing true cost per query. For AI workloads, this approach helps identify inefficient models or redundant invocations that inflate spend.</p><p>Inference costs are especially sensitive to query complexity and model selection.</p><p>Reduce data scanned to answer AI prompts that have frequently used data chunks that are flattened and stored in separate columns. This takes advantage of Snowflake’s natural columnar storage optimizations.</p><h4>Complexity</h4><p>Complex AI models with high token counts or large context windows consume more credits per inference. Teams should evaluate model efficiency, use caching where possible, and avoid overusing AI for simple tasks.</p><h4>Cost Anomalies Feature</h4><p>Snowflake’s Cost Anomalies feature (released May 2025) can flag unexpected spikes in AI usage, enabling proactive intervention.</p><h4>Query Tags</h4><p>Additionally, leveraging query tags on AI queries allows for real-time cost tracking and automated alerts when thresholds are breached.</p><h4>Resource Monitors</h4><p>Proactive cost optimization involves continuous monitoring and automation.</p><p>Resource monitors in Snowflake are designed to track and control credit consumption primarily for virtual warehouses and cloud services and are limited in their ability to monitor and limit AI Services costs.</p><p>They can only indirectly control the compute costs associated with executing AI functions by setting credit quotas and triggering alerts or suspensions when usage thresholds are exceeded on virtual warehouses where AI queries run.</p><p>You can assign a warehouse used for AI workloads to a resource monitor to prevent runaway compute spend, even though the AI token processing cost (serverless) is unaffected. For direct AI cost control, use budgets with monitoring on AI_SERVICES usage.</p><h4>Cost Insights and Cloud Services Optimizations Features</h4><p>Snowflake’s Cost Insights and Cloud Services Optimizations features provide recommendations to reduce inefficiencies, such as unused AI assets or poorly optimized queries.</p><h4>External Tools utilizing Account Usage Views</h4><p>External tools like Atlan can enhance visibility by leveraging metadata from ACCOUNT_USAGE to detect unused tables, materialized views, or transient data used in AI pipelines. Regular cleanup of unused AI artifacts and temporary data reduces both storage and compute costs.</p><h3>AI Usage Query History Views + AI Query Tags</h3><p>There are two new usage query history view for AI that show the credits consumed every time you interact with Cortex Agents or Snowflake Intelligence with granular token credits assigned even when a request results in one or more calls to underlying tools.</p><p>Both of the following views have a TOKENS_GRANULAR and a CREDITS_GRANULAR array in json that consist of nested values for each child object that is executed.</p><p>There are also new AI Query Tags called AGENT_TAGS and SNOWFLAKE_INTELLIGENCE_TAGS.</p><p>Both usage history views have the AGENT_TAGS column. These columns have a nested json array of tags that apply for each child object that is executed, if applicable.</p><p>USER_TAGS column also has an array in json showing the nested tags that apply.</p><blockquote><strong>CORTEX_AGENT_USAGE_HISTORY</strong></blockquote><ul><li>USER_TAGS</li><li>AGENT_TAGS</li><li>TOKENS_GRANULAR</li><li>CREDITS_GRANULAR</li></ul><pre>SELECT *<br>  FROM SNOWFLAKE.ACCOUNT_USAGE.CORTEX_AGENT_USAGE_HISTORY;</pre><blockquote><strong>SNOWFLAKE_INTELLIGENCE_USAGE_HISTORY</strong></blockquote><ul><li>USER_TAGS</li><li>AGENT_TAGS</li><li>SNOWFLAKE_INTELLIGENCE_TAGS</li><li>TOKENS_GRANULAR</li><li>CREDITS_GRANULAR</li></ul><pre>SELECT *<br>  FROM SNOWFLAKE.ACCOUNT_USAGE.SNOWFLAKE_INTELLIGENCE_USAGE_HISTORY;</pre><h3>Cost Summary</h3><p>To query standard compute costs for AI workloads in Snowflake, focus on the WAREHOUSE_METERING_HISTORY view in the SNOWFLAKE.ACCOUNT_USAGE schema, which tracks credit consumption for virtual warehouses used during AI operations such as running Cortex Analyst queries, refreshing Cortex Search indexes, or executing embedding jobs.</p><p>Since AI functions like AI_EXTRACT or AI_EMBED rely on standard warehouses for orchestration, even though the AI processing itself uses AI Services compute, their associated compute costs appear under CREDITS_USED_COMPUTE.</p><p>So, filter by WAREHOUSE_NAME and START_TIME to isolate usage during AI-related workloads and join with QUERY_HISTORY to attribute costs to specific AI queries, such as those using SNOWFLAKE.CORTEX functions.</p><p>Use resource monitors or tagging to allocate these costs by team or project, ensuring visibility into AI-driven compute spend alongside general usage.</p><h3>Cortex Code Costs</h3><p>Cortex Code (Coco) charges standard Cloud Services compute costs for accessing metadata in Snowflake such as using DESCRIBEor querying SNOWFLAKE.ACCOUNT_USAGEtables and these costs can be found in the regular QUERY_HISTORYview.</p><p>Additionally, Coco is charging costs as AI Services for the LLMs that are executed.</p><p>I love the fact that you can now select different versions of LLMs down in the lower right of the Cortex Code panel in the UI. See the chart for various costs per LLM in the Cortex Code CLI Costs section below.</p><p>These costs show under AI_SERVICES inWAREHOUSE_METERING_HISTORY in the following query:</p><pre>SELECT<br>  USAGE_DATE,<br>  SERVICE_TYPE,<br>  CREDITS_USED,<br>  CREDITS_USED * &lt;your_credit_rate&gt; AS estimated_cost_usd<br>FROM SNOWFLAKE.ORGANIZATION_USAGE.METERING_DAILY_HISTORY<br>WHERE SERVICE_TYPE IN (&#39;AI_SERVICES&#39;, &#39;AI_INFERENCE&#39;)<br>  AND USAGE_DATE &gt;= DATEADD(&#39;month&#39;, -3, CURRENT_DATE())<br>ORDER BY USAGE_DATE DESC;</pre><p>Please note that a Snowflake Trial Account has a limit of 10 credits per day for Coco use.</p><p>With the release of costing for Cortex Code UI and Cortex Code CLI, Snowflake has provided two usage history views: CORTEX_CODE_SNOWSIGHT_USAGE_HISTORY and CORTEX_CODE_CLI_USAGE_HISTORY. There is a 45-minute to 2-hour latency for costs to appear.</p><blockquote>What Does Coco Bill For</blockquote><p>In Coco’s own words -</p><p>1. Conversation text = token credits. Every message I generate (and every message you send that I process) counts as tokens billed to CORTEX_CODE_SNOWSIGHT_USAGE_HISTORY. The long explanations, the SQL I wrote out, all of it. You can see the per-model breakdown in the TOKENS_GRANULAR column — it shows cache_read_input, cache_write_input, input, and output tokens, each with different credit rates.</p><p>2. The SQL queries I executed = compute credits (cloud services). When I ran queries like DESCRIBE TABLE, SELECT ... FROM ACCOUNT_USAGE, those hit your warehouse (COMPUTE_WH) and incur cloud services credits, tracked in QUERY_HISTORY. These are tagged with {&quot;app&quot;:&quot;cortex_code_snowsight&quot;}. These are not token credits — they&#39;re standard warehouse compute costs, though most of these small metadata queries only cost fractions of a cent in cloud services credits.</p><p>In summary:</p><ul><li>My text/reasoning → token credits (Cortex Code pricing, ~$0.03–0.06 per exchange based on your data)</li><li>SQL I execute on your behalf → cloud services/compute credits (warehouse pricing, typically tiny for these queries)</li><li>Both show up in the rewritten query — token costs from CORTEX_CODE_SNOWSIGHT_USAGE_HISTORY and compute costs from QUERY_HISTORY</li></ul><blockquote>Plan Mode Costs</blockquote><p>Plan mode uses fewer tokens because -</p><ul><li>Input tokens: Each exchange re-sends the full conversation history. Plan mode encourages fewer, more targeted exchanges, so you accumulate less context re-processing.</li><li>Output tokens: Plan mode produces a concise plan first rather than immediately generating long code/explanations. You approve before the heavy work begins.</li></ul><p>However, the savings are modest in practice. Looking at your TOKENS_GRANULAR data, the biggest cost driver is cache_read_input (re-reading conversation context) and cache_write_input (new context). The output tokens (my actual responses) are a smaller fraction but have a higher per-token credit rate.</p><p>The most impactful way to reduce costs is shorter conversations — once a conversation gets long, every new exchange re-processes the full history.</p><p><strong><em>Starting a fresh conversation for a new topic is cheaper than continuing a long one.</em></strong></p><blockquote>Cortex Code Cost Limits</blockquote><p>Administrators can put limits on Cortex Code on a per user basis. See this article for details:</p><p><a href="https://docs.snowflake.com/en/user-guide/cortex-code/credit-usage-limit">Cost controls for Cortex Code | Snowflake Documentation</a></p><h3>Cortex Code UI Costs</h3><p>I’ve put together a cost query that is found in <strong>Appendix IV </strong>that queries across the UI and CLI as the primary CTE section, then arrives at details in a final query section.</p><p>The query gathers Cortex Code costs using the two new usage history views, and then gets the list of models as an on-row value for the final query.</p><p>The two new Cortex Code usage history views do not have a query id that can tie to query_history to view details about compute usage, but may be available in future enhancements.</p><p>For now, in order to join to query_history to get compute details, the request_id in the system query_tag, that Snowflake applies to all Cortex Code aligns on the request_id and only shows for requests that incurred compute. However, since a single conversation prompt could result in multiple queries, this still would return repeated compute detail rows so it not ideal. The second query in <strong>Appendix IV</strong> links to query_history and aggregates the multiple rows in query_history to a single request_id using average for some columns.</p><p><strong><em>However, using a new session in the UI / CLI per new conversation can provide finer grain cost alignment to compute details.</em></strong></p><h3>Cortex Code CLI Costs</h3><p>Here is a chart that covers Cortex Code costs that is reflected in the Snowflake Service Consumption Table document.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/720/1*RAJjUGIGHZMyQwx5YI4AvA.png" /></figure><p><a href="https://www.snowflake.com/legal-files/CreditConsumptionTable.pdf">Cortex Code CLI Pricing(Source: Snowflake)</a></p><h3>Cortex Search Services Cost</h3><p>The Cortex Search Service incurs “Serving” compute costs as daily charges even when not actively used.</p><p>You can run this command to see which Cortex Search Services are running:</p><blockquote>SHOW CORTEX SEARCH SERVICES IN ACCOUNT;</blockquote><p>Unlike virtual warehouses that can auto-suspend, the serving layer of a Cortex Search service runs continuously to enable low-latency queries. You pay a fee based on the uncompressed indexed data size (in GB/month), which includes both your source data and vector embeddings, regardless of whether any queries are executed.</p><p>This is a “running cost” that persists as long as the service’s SERVING status is RUNNING. For example:</p><p>A very small index &lt; 1 GB ~ less than 1/1000th of a credit/month.<br>A 50GB index incurs ~315 credits/month (at $3/credit).<br>A 100GB index costs ~630 credits/month (~$1,890), even with zero queries.</p><p>Key Insight: The cost is tied to availability, not usage. To avoid these charges, suspend the service when not in use (e.g., during development). Resuming takes minutes, and it’s a best practice to only keep production services running.</p><blockquote>ALTER CORTEX SEARCH SERVICE my_search_service SUSPEND SERVING;</blockquote><p>Replace my_search_service in this statement with your service’s name. This halts billing for the serving layer while preserving your index. To resume later, use RESUME SERVING. You need the OPERATE privilege on the service to perform this action.</p><h3>Conclusion</h3><blockquote>In conclusion, mastering AI cost querying in Snowflake requires a blend of technical precision and strategic governance.</blockquote><p>Focus on accurate cost attribution, optimized query patterns, smart resource sizing, and vigilance on egress.</p><p>Use native tools like ACCOUNT_USAGE, QUERY_ATTRIBUTION_HISTORY, and WAREHOUSE_METERING_HISTORY to build a transparent cost model.</p><p>Combine this with tagging, budgets, and anomaly detection to turn AI from a cost center into a value-driven capability.</p><p>With the right approach, Snowflake’s AI Data Cloud can deliver powerful insights without breaking the bank.</p><h3>APPENDIX I — Invoice Cost Details</h3><p>The COST column in this query should equal the amount of AI INFERENCE and AI SERVICES line-items on the Snowflake Monthly Statement.</p><p>A COMPUTE aggregation is also provided as a total for virtual warehouse cost.</p><pre>WITH invoice_cost_details as (<br><br>    -- COMPUTE - BY WAREHOUSE OR USER NAME<br>    SELECT start_date as USAGE_DATE,<br>        &#39;COMPUTE&#39; as SERVICE_GROUP,<br>        &#39;COMPUTE_WAREHOUSES&#39; AS SERVICE_NAME,<br>        SUM(CREDITS_USED_CLOUD_SERVICES) AS CREDITS_USED,<br>        SUM(CREDITS_USED_CLOUD_SERVICES) * &lt;&lt; your credit price here &gt;&gt; AS COST,<br>        0 as TOKENS<br>    FROM snowflake.account_usage.query_history<br>    WHERE TO_DATE(START_DATE) &gt;= DATE_TRUNC(&#39;MONTH&#39;,DATEADD(day, -60, CURRENT_DATE))<br>    AND WAREHOUSE_NAME = &#39;&lt;&lt; your warehouse name here &gt;&gt;&#39;<br>    GROUP BY ALL<br>    UNION ALL  <br>    <br>    <br>    -- COREX AISQL<br>    SELECT<br>        TO_DATE(USAGE_TIME) AS USAGE_DATE,<br>        &#39;AI SERVICES&#39; as SERVICE_GROUP,<br>        &#39;CORTEX_AISQL&#39; AS SERVICE_NAME,<br>        SUM(TOKEN_CREDITS) AS CREDITS_USED,<br>        SUM(TOKEN_CREDITS) * &lt;&lt; your credit price here &gt;&gt; AS COST,<br>        SUM(TOKENS) as TOKENS<br>    FROM SNOWFLAKE.ACCOUNT_USAGE.CORTEX_AISQL_USAGE_HISTORY<br>    WHERE TO_DATE(USAGE_TIME) &gt;= DATE_TRUNC(&#39;MONTH&#39;,DATEADD(day, -60, CURRENT_DATE))<br>    GROUP BY TO_DATE(USAGE_TIME), DATABASE_NAME<br>    UNION ALL<br>    <br>    <br>    -- DOCUMENT AI (sequence out of alpha order but placed near other doc ai charges)<br>    SELECT<br>        TO_DATE(START_TIME) AS USAGE_DATE,<br>        &#39;INFERENCE DETAIL&#39; as SERVICE_GROUP,<br>        &#39;DOCUMENT_AI&#39; AS SERVICE_NAME,<br>        SUM(CREDITS_USED) AS CREDITS_USED,<br>        SUM(CREDITS_USED) * &lt;&lt; your credit price here &gt;&gt; AS COST,<br>        0 as TOKENS<br>    FROM SNOWFLAKE.ACCOUNT_USAGE.DOCUMENT_AI_USAGE_HISTORY<br>    WHERE TO_DATE(START_TIME) &gt;= DATE_TRUNC(&#39;MONTH&#39;,DATEADD(day, -60, CURRENT_DATE))<br>    GROUP BY TO_DATE(START_TIME), DATABASE_NAME<br>    UNION ALL<br>    <br>    -- CORTEX FINE_TUNING<br>    SELECT<br>        TO_DATE(START_TIME) AS USAGE_DATE,<br>        &#39;AI SERVICES&#39; as SERVICE_GROUP,<br>        &#39;CORTEX_FINE_TUNING&#39; AS SERVICE_NAME,<br>        SUM(TOKEN_CREDITS) AS CREDITS_USED,<br>        SUM(TOKEN_CREDITS) * &lt;&lt; your credit price here &gt;&gt; AS COST,<br>        SUM(TOKENS) as TOKENS<br>    FROM SNOWFLAKE.ACCOUNT_USAGE.CORTEX_FINE_TUNING_USAGE_HISTORY<br>    WHERE TO_DATE(START_TIME) &gt;= DATE_TRUNC(&#39;MONTH&#39;,DATEADD(day, -60, CURRENT_DATE))<br>    GROUP BY TO_DATE(START_TIME), DATABASE_NAME<br>    UNION ALL<br>    <br>    -- CORTEX FUNCTIONS USAGE<br>    SELECT<br>        TO_DATE(START_TIME) AS USAGE_DATE,<br>        &#39;AI SERVICES&#39; as SERVICE_GROUP,<br>        &#39;CORTEX_FUNCTIONS_USAGE&#39; AS SERVICE_NAME,<br>        SUM(TOKEN_CREDITS) AS CREDITS_USED,<br>        SUM(TOKEN_CREDITS) * &lt;&lt; your credit price here &gt;&gt; AS COST,<br>        SUM(TOKENS) as TOKENS<br>    FROM SNOWFLAKE.ACCOUNT_USAGE.CORTEX_FUNCTIONS_USAGE_HISTORY<br>    WHERE TO_DATE(START_TIME) &gt;= DATE_TRUNC(&#39;MONTH&#39;,DATEADD(day, -60, CURRENT_DATE))<br>    GROUP BY TO_DATE(START_TIME), DATABASE_NAME<br>    UNION ALL<br>    <br>    -- CORTEX PROVISIONED<br>    SELECT<br>        TO_DATE(INTERVAL_START_TIME) AS USAGE_DATE,<br>        &#39;AI SERVICES&#39; as SERVICE_GROUP,<br>        &#39;CORTEX_AISQL&#39; AS SERVICE_NAME,<br>        SUM(PTU_CREDITS) AS CREDITS_USED,<br>        SUM(PTU_CREDITS) * &lt;&lt; your credit price here &gt;&gt; AS COST,<br>        0 as TOKENS<br>    FROM SNOWFLAKE.ACCOUNT_USAGE.CORTEX_PROVISIONED_THROUGHPUT_USAGE_HISTORY<br>    WHERE TO_DATE(INTERVAL_START_TIME) &gt;= DATE_TRUNC(&#39;MONTH&#39;,DATEADD(day, -60, CURRENT_DATE))<br>    GROUP BY TO_DATE(INTERVAL_START_TIME), DATABASE_NAME<br>    UNION ALL<br>    <br>    -- REST API ACTIVITY<br>    SELECT<br>        TO_DATE(START_TIME) AS USAGE_DATE,<br>        &#39;AI SERVICES&#39; as SERVICE_GROUP,<br>        &#39;CORTEX_REST_API&#39; AS SERVICE_NAME,<br>        0 AS CREDITS_USED,<br>        SUM(0) * &lt;&lt; your credit price here &gt;&gt; AS COST,<br>        SUM(TOKENS) as TOKENS<br>    FROM SNOWFLAKE.ACCOUNT_USAGE.CORTEX_REST_API_USAGE_HISTORY<br>    WHERE TO_DATE(START_TIME) &gt;= DATE_TRUNC(&#39;MONTH&#39;,DATEADD(day, -60, CURRENT_DATE))<br>    GROUP BY TO_DATE(START_TIME), DATABASE_NAME<br>    UNION ALL<br>    <br>    -- CORTEX SEARCH (used for serving/text embeddings)<br>    SELECT<br>        TO_DATE(USAGE_DATE) AS USAGE_DATE,<br>        &#39;AI SERVICES&#39; as SERVICE_GROUP,<br>        &#39;CORTEX_SEARCH_DAILY&#39; AS SERVICE_NAME,<br>        SUM(CREDITS) AS CREDITS_USED,<br>        SUM(CREDITS) * &lt;&lt; your credit price here &gt;&gt; AS COST,<br>        SUM(TOKENS) as TOKENS<br>    FROM SNOWFLAKE.ACCOUNT_USAGE.CORTEX_SEARCH_DAILY_USAGE_HISTORY<br>    WHERE TO_DATE(USAGE_DATE) &gt;= DATE_TRUNC(&#39;MONTH&#39;,DATEADD(day, -60, CURRENT_DATE))<br>    GROUP BY TO_DATE(USAGE_DATE), DATABASE_NAME<br><br>)<br>SELECT<br>    USAGE_DATE,<br>    SERVICE_GROUP,  -- AI INFERENCE, AI SERVICES or COMPUTE<br>    SERVICE_NAME,<br>    CREDITS_USED,<br>    COST, -- Adjust multiplier as needed<br>    TOKENS<br>FROM invoice_cost_details<br>ORDER BY <br>;</pre><h3>APPENDIX II— Individual AI Query Details</h3><p>Any of the Cortex Usage History views that return a Query ID can be applied to the join and query fields in this query.</p><pre>SELECT <br>    qh.start_time::date as start_date<br>    , &#39;aisql_usage&#39; as cost_category<br>    , c.query_id<br>    , qh.warehouse_name<br>    , qh.user_name<br>    , c.model_name<br>    , c.function_name<br>    , c.token_credits<br>    , c.token_credits * &lt;&lt; your credit price here &gt;&gt; as cost<br>    , c.tokens<br>    , qh.query_text<br>    , qh.database_name<br>    , qh.schema_name<br>    , qh.query_type<br>    , qh.role_name<br>    , qh.query_tag<br>    , qh.start_time<br>    , qh.end_time<br>    , round(qh.total_elapsed_time / (1000*60),2) duration_mins<br>    , qh.total_elapsed_time as duration_as_ms<br>    , round(div0null(qh.bytes_scanned,(pow(1024,3))),1) as bytes_scan_gb<br>    , round(div0null(qh.bytes_spilled_to_local_storage,pow(1024,3)),2) as local_spill_gb<br>    , round(div0null(qh.bytes_spilled_to_remote_storage,pow(1024,3)),2) as remote_spill_gb<br>    , round(qh.percentage_scanned_from_cache,2) as cache_pct<br>    , round(div0null(qh.bytes_sent_over_the_network,pow(1024,3)),2) as network_gb<br>    , qh.queued_overload_time<br>    , qh.queued_provisioning_time<br>    , qh.transaction_blocked_time<br>    , qh.rows_produced<br>    , qh.rows_inserted<br>    , qh.rows_updated<br>    , qh.rows_deleted<br>    , qh.rows_unloaded<br>    , qh.external_function_total_invocations<br>    , qh.external_function_total_sent_bytes<br>    , qh.external_function_total_sent_rows<br>    , qh.external_function_total_received_bytes<br>    , qh.external_function_total_received_rows<br>    , qh.rows_written_to_result<br>FROM SNOWFLAKE.ACCOUNT_USAGE.QUERY_HISTORY qh<br>JOIN SNOWFLAKE.ACCOUNT_USAGE.CORTEX_AISQL_USAGE_HISTORY c<br>ON c.QUERY_ID = qh.QUERY_ID<br>WHERE TO_DATE(qh.START_TIME) &gt;= DATE_TRUNC(&#39;MONTH&#39;,DATEADD(day, -60, CURRENT_DATE))<br>;</pre><h3>APPENDIX III — Second-Level Cost Details</h3><p>These second-level costs encompass all of the other Cortex Usage History views at the time of the writing of this article and were collected to be able to see the lower layers of activities in Snowflake where Cortex costs are being attributed for tracking purposes.</p><pre>WITH second_level_details as (<br><br>    -- CORTEX ANALYST<br>    SELECT<br>        TO_DATE(START_TIME) AS USAGE_DATE,<br>        &#39;SECOND_LEVEL_DETAIL&#39; as SERVICE_GROUP,<br>        &#39;CORTEX_ANALYST&#39; AS SERVICE_NAME,<br>        &#39;N/A&#39; as DATABASE_NAME,<br>        SUM(CREDITS) AS CREDITS_USED,<br>        SUM(CREDITS) * &lt;&lt; your credit price here &gt;&gt; AS COST,<br>        0 as TOKENS<br>    FROM SNOWFLAKE.ACCOUNT_USAGE.CORTEX_ANALYST_USAGE_HISTORY<br>    WHERE TO_DATE(START_TIME) &gt;= DATE_TRUNC(&#39;MONTH&#39;,DATEADD(day, -60, CURRENT_DATE))<br>    GROUP BY TO_DATE(START_TIME), DATABASE_NAME<br>    UNION ALL<br>    <br>    --CORTEX DOCUMENT PROCESSING<br>    SELECT<br>        TO_DATE(START_TIME) AS USAGE_DATE,<br>        &#39;SECOND_LEVEL_DETAIL&#39; as SERVICE_GROUP,<br>        &#39;DOCUMENT_PROCESSING&#39; AS SERVICE_NAME,<br>        &#39;N/A&#39; as DATABASE_NAME,<br>        SUM(CREDITS_USED) AS CREDITS_USED,<br>        SUM(CREDITS_USED) * &lt;&lt; your credit price here &gt;&gt; AS COST,<br>        0 as TOKENS<br>    FROM SNOWFLAKE.ACCOUNT_USAGE.CORTEX_DOCUMENT_PROCESSING_USAGE_HISTORY<br>    WHERE TO_DATE(START_TIME) &gt;= DATE_TRUNC(&#39;MONTH&#39;,DATEADD(day, -60, CURRENT_DATE))<br>    GROUP BY TO_DATE(START_TIME), DATABASE_NAME<br>    UNION ALL<br>    <br>    -- CORTEX FUNCTIONS USAGE<br>    SELECT<br>        TO_DATE(START_TIME) AS USAGE_DATE,<br>        &#39;SECOND_LEVEL_DETAIL&#39; as SERVICE_GROUP,<br>        &#39;CORTEX_FUNCTIONS_USAGE&#39; AS SERVICE_NAME,<br>        &#39;N/A&#39; as DATABASE_NAME,<br>        SUM(TOKEN_CREDITS) AS CREDITS_USED,<br>        SUM(TOKEN_CREDITS) * &lt;&lt; your credit price here &gt;&gt; AS COST,<br>        SUM(TOKENS) as TOKENS<br>    FROM SNOWFLAKE.ACCOUNT_USAGE.CORTEX_FUNCTIONS_USAGE_HISTORY<br>    WHERE TO_DATE(START_TIME) &gt;= DATE_TRUNC(&#39;MONTH&#39;,DATEADD(day, -60, CURRENT_DATE))<br>    GROUP BY TO_DATE(START_TIME), DATABASE_NAME<br>    UNION ALL<br>    <br>    <br>    -- CORTEX PROVISIONED<br>    SELECT<br>        TO_DATE(INTERVAL_START_TIME) AS USAGE_DATE,<br>        &#39;SECOND_LEVEL_DETAIL&#39; as SERVICE_GROUP,<br>        &#39;CORTEX_AISQL&#39; AS SERVICE_NAME,<br>        &#39;N/A&#39; as DATABASE_NAME,<br>        SUM(PTU_CREDITS) AS CREDITS_USED,<br>        SUM(PTU_CREDITS) * &lt;&lt; your credit price here &gt;&gt; AS COST,<br>        0 as TOKENS<br>    FROM SNOWFLAKE.ACCOUNT_USAGE.CORTEX_PROVISIONED_THROUGHPUT_USAGE_HISTORY<br>    WHERE TO_DATE(INTERVAL_START_TIME) &gt;= DATE_TRUNC(&#39;MONTH&#39;,DATEADD(day, -60, CURRENT_DATE))<br>    GROUP BY TO_DATE(INTERVAL_START_TIME), DATABASE_NAME<br>    UNION ALL<br>    <br>    -- CORTEX SEARCH SERVICE<br>    SELECT<br>        TO_DATE(START_TIME) AS USAGE_DATE,<br>        &#39;SECOND_LEVEL_DETAIL&#39; as SERVICE_GROUP,<br>        &#39;CORTEX_SEARCH_SERVING&#39; AS SERVICE_NAME,<br>        DATABASE_NAME,<br>        SUM(CREDITS) AS CREDITS_USED,<br>        SUM(CREDITS) * &lt;&lt; your credit price here &gt;&gt; AS COST,<br>        0 as TOKENS<br>    FROM SNOWFLAKE.ACCOUNT_USAGE.CORTEX_SEARCH_SERVING_USAGE_HISTORY<br>    WHERE TO_DATE(START_TIME) &gt;= DATE_TRUNC(&#39;MONTH&#39;,DATEADD(day, -60, CURRENT_DATE))<br>    GROUP BY TO_DATE(START_TIME), DATABASE_NAME<br>    <br>)<br>SELECT<br>    USAGE_DATE,<br>    SERVICE_GROUP,  -- SECOND_LEVEL_DETAIL<br>    SERVICE_NAME,<br>    CREDITS_USED,<br>    COST, -- Adjust multiplier as needed<br>    TOKENS<br>FROM second_level_details<br>ORDER BY USAGE_DATE DESC<br>;</pre><h3>APPENDIX IV — Cortex Code Cost Query</h3><p>This query uses CORTEX_CODE_SNOWSIGHT_USAGE_HISTORY and CORTEX_CODE_CLI_USAGE_HISTORY views as the primary CTE section.</p><p>It then gathers a list of models that were used in a list for on-row display.</p><pre>WITH all_cortex_code_usage AS (<br>    SELECT &#39;snowsight_ui&#39; AS source, USER_NAME, REQUEST_ID, PARENT_REQUEST_ID, USAGE_TIME,<br>        TOKEN_CREDITS, TOKENS, TOKENS_GRANULAR, CREDITS_GRANULAR<br>    FROM SNOWFLAKE.ACCOUNT_USAGE.CORTEX_CODE_SNOWSIGHT_USAGE_HISTORY<br>    WHERE USAGE_TIME &gt;= DATEADD(&#39;day&#39;, -30, CURRENT_TIMESTAMP())<br>    UNION ALL<br>    SELECT &#39;cli&#39; AS source, USER_NAME, REQUEST_ID, NULL AS PARENT_REQUEST_ID, USAGE_TIME,<br>        TOKEN_CREDITS, TOKENS, TOKENS_GRANULAR, CREDITS_GRANULAR<br>    FROM SNOWFLAKE.ACCOUNT_USAGE.CORTEX_CODE_CLI_USAGE_HISTORY<br>    WHERE USAGE_TIME &gt;= DATEADD(&#39;day&#39;, -30, CURRENT_TIMESTAMP())<br>),<br>cortex_code_summary AS (<br>    SELECT<br>        REQUEST_ID,<br>        SOURCE,<br>        USER_NAME,<br>        PARENT_REQUEST_ID,<br>        USAGE_TIME,<br>        TOKEN_CREDITS,<br>        TOKEN_CREDITS * &lt;&lt;your token cost here&gt;&gt; AS cortex_code_cost_usd,<br>        TOKENS,<br>        TOKENS_GRANULAR,<br>        CREDITS_GRANULAR,<br>        LISTAGG(DISTINCT f.key, &#39;, &#39;) WITHIN GROUP (ORDER BY f.key) AS models_used<br>    FROM all_cortex_code_usage,<br>        LATERAL FLATTEN(TOKENS_GRANULAR) f<br>    GROUP BY REQUEST_ID, SOURCE, USER_NAME, PARENT_REQUEST_ID, USAGE_TIME,<br>        TOKEN_CREDITS, TOKENS, TOKENS_GRANULAR, CREDITS_GRANULAR<br>)<br>SELECT<br>    REQUEST_ID,<br>    SOURCE AS cortex_code_source,<br>    USER_NAME,<br>    USAGE_TIME,<br>    MODELS_USED,<br>    TOKENS AS cortex_code_tokens,<br>    TOKEN_CREDITS AS cortex_code_credits,<br>    cortex_code_cost_usd,<br>    TOKENS_GRANULAR AS cortex_code_tokens_granular,<br>    CREDITS_GRANULAR AS cortex_code_credits_granular<br>FROM cortex_code_summary<br>ORDER BY TOKEN_CREDITS DESC;</pre><p>This query joins to query_history on request_id, however, there can be two query_ids to each request_id so the compute metrics have been aggregated per request.</p><pre>WITH all_cortex_code_usage AS (<br>    SELECT &#39;snowsight_ui&#39; AS source, USER_NAME, REQUEST_ID, PARENT_REQUEST_ID, USAGE_TIME,<br>        TOKEN_CREDITS, TOKENS, TOKENS_GRANULAR, CREDITS_GRANULAR<br>    FROM SNOWFLAKE.ACCOUNT_USAGE.CORTEX_CODE_SNOWSIGHT_USAGE_HISTORY<br>    WHERE USAGE_TIME &gt;= DATEADD(&#39;day&#39;, -30, CURRENT_TIMESTAMP())<br>    UNION ALL<br>    SELECT &#39;cli&#39; AS source, USER_NAME, REQUEST_ID, NULL AS PARENT_REQUEST_ID, USAGE_TIME,<br>        TOKEN_CREDITS, TOKENS, TOKENS_GRANULAR, CREDITS_GRANULAR<br>    FROM SNOWFLAKE.ACCOUNT_USAGE.CORTEX_CODE_CLI_USAGE_HISTORY<br>    WHERE USAGE_TIME &gt;= DATEADD(&#39;day&#39;, -30, CURRENT_TIMESTAMP())<br>),<br>cortex_code_summary AS (<br>    SELECT<br>        REQUEST_ID,<br>        SOURCE,<br>        USER_NAME,<br>        PARENT_REQUEST_ID,<br>        USAGE_TIME,<br>        TOKEN_CREDITS,<br>        TOKEN_CREDITS * 3.83 AS cortex_code_cost_usd,<br>        TOKENS,<br>        TOKENS_GRANULAR,<br>        CREDITS_GRANULAR,<br>        LISTAGG(DISTINCT f.key, &#39;, &#39;) WITHIN GROUP (ORDER BY f.key) AS models_used<br>    FROM all_cortex_code_usage,<br>        LATERAL FLATTEN(TOKENS_GRANULAR) f<br>    GROUP BY REQUEST_ID, SOURCE, USER_NAME, PARENT_REQUEST_ID, USAGE_TIME,<br>        TOKEN_CREDITS, TOKENS, TOKENS_GRANULAR, CREDITS_GRANULAR<br>),<br>compute_by_request AS (<br>    SELECT<br>        REPLACE(QUERY_TAG, &#39;cortex-agent-&#39;, &#39;&#39;) AS REQUEST_ID,<br>        MAX(SESSION_ID) AS SESSION_ID,<br>        COUNT(*) AS compute_query_count,<br>        SUM(CREDITS_USED_CLOUD_SERVICES) AS cloud_services_credits,<br>        SUM(TOTAL_ELAPSED_TIME) AS total_elapsed_ms,<br>        SUM(BYTES_SCANNED) AS total_bytes_scanned,<br>        SUM(ROWS_PRODUCED) AS total_rows_produced,<br>        LISTAGG(DISTINCT WAREHOUSE_NAME, &#39;, &#39;) AS warehouses_used,<br>        LISTAGG(DISTINCT QUERY_TYPE, &#39;, &#39;) AS query_types,<br>        LISTAGG(DISTINCT DATABASE_NAME, &#39;, &#39;) AS databases_used,<br>        LISTAGG(DISTINCT SCHEMA_NAME, &#39;, &#39;) AS schemas_used,<br>        LISTAGG(DISTINCT ROLE_NAME, &#39;, &#39;) AS roles_used,<br>        MIN(START_TIME) AS earliest_start_time,<br>        MAX(END_TIME) AS latest_end_time,<br>        ROUND(AVG(DIV0NULL(BYTES_SPILLED_TO_LOCAL_STORAGE, POW(1024,3))), 2) AS avg_local_spill_gb,<br>        ROUND(AVG(DIV0NULL(BYTES_SPILLED_TO_REMOTE_STORAGE, POW(1024,3))), 2) AS avg_remote_spill_gb,<br>        ROUND(AVG(PERCENTAGE_SCANNED_FROM_CACHE), 2) AS avg_cache_pct,<br>        ROUND(AVG(DIV0NULL(BYTES_SENT_OVER_THE_NETWORK, POW(1024,3))), 2) AS avg_network_gb,<br>        AVG(QUEUED_OVERLOAD_TIME) AS avg_queued_overload_time,<br>        AVG(QUEUED_PROVISIONING_TIME) AS avg_queued_provisioning_time,<br>        AVG(TRANSACTION_BLOCKED_TIME) AS avg_transaction_blocked_time,<br>        SUM(ROWS_INSERTED) AS total_rows_inserted,<br>        SUM(ROWS_UPDATED) AS total_rows_updated,<br>        SUM(ROWS_DELETED) AS total_rows_deleted,<br>        SUM(ROWS_UNLOADED) AS total_rows_unloaded,<br>        SUM(EXTERNAL_FUNCTION_TOTAL_INVOCATIONS) AS total_ext_func_invocations,<br>        SUM(EXTERNAL_FUNCTION_TOTAL_SENT_BYTES) AS total_ext_func_sent_bytes,<br>        SUM(EXTERNAL_FUNCTION_TOTAL_SENT_ROWS) AS total_ext_func_sent_rows,<br>        SUM(EXTERNAL_FUNCTION_TOTAL_RECEIVED_BYTES) AS total_ext_func_received_bytes,<br>        SUM(EXTERNAL_FUNCTION_TOTAL_RECEIVED_ROWS) AS total_ext_func_received_rows,<br>        SUM(ROWS_WRITTEN_TO_RESULT) AS total_rows_written_to_result<br>    FROM SNOWFLAKE.ACCOUNT_USAGE.QUERY_HISTORY<br>    WHERE QUERY_TAG LIKE &#39;cortex-agent-%&#39;<br>        AND START_TIME &gt;= DATEADD(&#39;day&#39;, -30, CURRENT_TIMESTAMP())<br>    GROUP BY REPLACE(QUERY_TAG, &#39;cortex-agent-&#39;, &#39;&#39;)<br>)<br>SELECT<br>    cc.REQUEST_ID,<br>    cq.SESSION_ID AS compute_session_id,<br>    cc.SOURCE AS cortex_code_source,<br>    cc.USER_NAME,<br>    cc.USAGE_TIME,<br>    cc.MODELS_USED,<br>    cc.TOKENS AS cortex_code_tokens,<br>    cc.TOKEN_CREDITS AS cortex_code_credits,<br>    cc.cortex_code_cost_usd,<br>    cc.TOKENS_GRANULAR AS cortex_code_tokens_granular,<br>    cc.CREDITS_GRANULAR AS cortex_code_credits_granular,<br>    cq.compute_query_count,<br>    cq.warehouses_used,<br>    cq.query_types,<br>    cq.databases_used,<br>    cq.schemas_used,<br>    cq.roles_used,<br>    cq.cloud_services_credits,<br>    COALESCE(cq.cloud_services_credits, 0) * 3.83 AS compute_cost_usd,<br>    cq.earliest_start_time,<br>    cq.latest_end_time,<br>    ROUND(DIV0NULL(cq.total_elapsed_ms, (1000*60)), 2) AS total_duration_mins,<br>    ROUND(DIV0NULL(cq.total_bytes_scanned, POW(1024,3)), 3) AS total_scan_gb,<br>    cq.total_rows_produced,<br>    (cc.TOKEN_CREDITS + COALESCE(cq.cloud_services_credits, 0)) AS total_credits,<br>    (cc.TOKEN_CREDITS + COALESCE(cq.cloud_services_credits, 0)) * 3.83 AS total_cost_usd,<br>    cq.avg_local_spill_gb,<br>    cq.avg_remote_spill_gb,<br>    cq.avg_cache_pct,<br>    cq.avg_network_gb,<br>    cq.avg_queued_overload_time,<br>    cq.avg_queued_provisioning_time,<br>    cq.avg_transaction_blocked_time,<br>    cq.total_rows_inserted,<br>    cq.total_rows_updated,<br>    cq.total_rows_deleted,<br>    cq.total_rows_unloaded,<br>    cq.total_ext_func_invocations,<br>    cq.total_ext_func_sent_bytes,<br>    cq.total_ext_func_sent_rows,<br>    cq.total_ext_func_received_bytes,<br>    cq.total_ext_func_received_rows,<br>    cq.total_rows_written_to_result<br>FROM cortex_code_summary cc<br>LEFT JOIN compute_by_request cq ON cq.REQUEST_ID = cc.REQUEST_ID<br>ORDER BY cc.TOKEN_CREDITS DESC;<br><br></pre><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=07331811d42d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI-Ready Data — the First Step in the AI Journey to unlocking Intelligent Insights]]></title>
            <link>https://medium.com/@angelamarieharney/ai-ready-data-the-first-step-in-the-ai-journey-to-unlocking-intelligent-insights-346601c73b20?source=rss-7e068c9737ea------2</link>
            <guid isPermaLink="false">https://medium.com/p/346601c73b20</guid>
            <dc:creator><![CDATA[Angela Harney]]></dc:creator>
            <pubDate>Mon, 15 Dec 2025 00:34:05 GMT</pubDate>
            <atom:updated>2025-12-15T00:34:05.379Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>AI-Ready Data — the First Step </strong>in the AI Journey to unlocking Intelligent Insights</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/780/1*6Qs18edpxT6CwvMKmel18w.png" /></figure><h3>Definition</h3><p><strong><em>AI-ready data is information that is meticulously cleaned, structured, and enriched with context, ensuring that it can be efficiently processed by AI models with minimal friction so they can be reliable.</em></strong></p><p>Unlike traditional “<strong>good enough data</strong>,” which focuses on accuracy and completeness, <strong>AI-ready data goes further</strong> by incorporating rich metadata, semantic meaning, and data lineage, enabling AI systems, especially large language models, to interpret, reason, and act on information with greater reliability and insight.</p><h3>Purpose</h3><p>In today’s rapidly evolving data-driven digital landscape, AI success depends on quality, not quantity. Without AI-Ready Data, even advanced algorithms fail to deliver accurate results. The success of AI systems hinges not on the sophistication of algorithms alone, but on the quality and readiness of the data they consume.</p><p>As organizations increasingly rely on AI for decision-making, from fraud detection to real-time market analysis, having data that is not only accurate and up-to-date, but also accessible and governed across distributed systems has become a strategic imperative.</p><p>Without this foundation, even the most advanced AI applications risk producing biased, outdated, or hallucinated outputs, undermining trust and effectiveness.</p><p>Embarking on an AI journey begins not with complex algorithms, but with the foundational steps of preparing <strong>AI-ready data</strong> that is clean, relevant, structured, and secure. High-quality data is the fuel that powers accurate and reliable AI models, making its preparation a critical success factor.</p><p><strong>Index</strong></p><p><a href="#_Toc216611664">Step 1 — Creating an Intelligence Layer of Data Architecture</a><br><a href="#_Toc216611665">Step 2 — Define Clear AI Objectives</a><br><a href="#_Toc216611666">Step 3 — Conduct a Data Audit</a><br><a href="#_Toc216611667">Step 4 — Establish Common Business Definitions</a><br><a href="#_Toc216611668">Step 5 — Cleanse and Standardize</a><br><a href="#_Toc216611669">Step 6 — Enrich and Integrate</a><br><a href="#_Toc216611670">Step 7 — Structure and Label for Learning</a><br><a href="#_Toc216611671">Step 8 — Ensure Security and Compliance</a><br><a href="#_Toc216611672">Step 9 — Automate Curated Data with Pipelines</a><br><a href="#_Toc216614794">Step 10 — Store Development, Training, Validation, Testing and Output Datasets in Intelligence Layer</a><br><a href="#_Toc216611678">Step 11 — Validate, Monitor and Certify Continuously</a><br><a href="#_Toc216611679">Conclusion</a></p><h3>Step 1 — Creating an Intelligence Layer of Data Architecture</h3><p>Artificial intelligence is transforming how we process information and make decisions, with the Intelligence Layer serving as a critical component in this evolution.</p><p>Design this layer over your existing semantic layer to serve AI tools and solutions allowing them to perceive, reason, and make optimal decisions by leveraging advanced capabilities in vision, speech recognition, natural language processing, and knowledge acquisition.</p><p>At its core, the Intelligence Layer relies on AI-ready data — <strong>data that is pre-cleaned, structured, and enriched with metadata to ensure high quality, relevance, and consistency</strong>, which is essential for training accurate and reliable AI models. Without such prepared data, even the most sophisticated algorithms would struggle to deliver meaningful insights or perform effectively in real-world applications.</p><ul><li><strong>The Intelligence Layer integrates perception, cognition, and decision-making capabilities, allowing systems to interpret complex inputs and respond intelligently</strong>.</li><li>AI-ready data is <strong>characterized by its readiness for immediate use in AI applications</strong>, differing from traditional data through superior structure, timeliness, and scalability.</li><li>Ensuring data quality and consistency is paramount, as biased or inaccurate data can lead to flawed outcomes, even in advanced AI systems.</li><li>Best practices for preparing AI-ready data include using consistent terminology, avoiding ambiguous pronouns, and structuring content with clear headings and bullet points to enhance machine readability.</li></ul><h3>Step 2 — Define Clear AI Objectives</h3><p>Identify the specific business problem your AI will solve. Whether it’s <strong>predictive analytics</strong>, <strong>image recognition</strong>, or <strong>natural language processing (NLP)</strong>, your goal dictates the type of data needed, such as historical records, labeled images, or text transcripts, etc.</p><p>Set up a <strong>metadata engine</strong> to align data with business goals by standardizing key definitions like “active customer” or “revenue.” It serves as a centralized source for data context around ownership, lineage, and rules to ensure consistency across teams and AI systems. This clarity improves governance and ensures accurate, actionable insights.</p><h3>Step 3 — Conduct a Data Audit</h3><p>Assess your existing data for <strong>completeness, consistency, accuracy, and freshness</strong>. This audit reveals gaps, duplicates, and inconsistencies, forming the baseline for improvement and ensuring only reliable data feeds your models.</p><h3>Step 4 — Establish Common Business Definitions</h3><p>Before data can be effectively used, align stakeholders on <strong>standardized business terms and metrics</strong>, such as “active customer,” “churn,” or “revenue.” Inconsistent definitions create data silos and model inaccuracies. Implement a <strong>business glossary</strong> within your data governance framework to ensure everyone, from data scientists to executives, interprets data the same way.</p><h3>Step 5 — Cleanse and Standardize</h3><p>Eliminate noise by <strong>removing duplicates, correcting errors, and filling missing values</strong>. Standardize formats, dates, units, naming conventions, across datasets to ensure uniformity and prevent model bias.</p><p>De-normalize data by combining sources, like purchases with customer sentiment, to create unified datasets to help reveal behavioral patterns, such as linking buying history with customer feedback to improve personalization and retention.</p><ul><li>Common pre-fixed and post-fixed abbreviations can be applied to table names in an Intelligence Layer of data for clarity. <strong>ML_</strong> can be used for tables, similar to using DIM_ or FACT_. Model tables can be prefixed with<strong> MODEL_</strong>. Output results can be post-fixed with <strong>_RESULTS </strong>or <strong>_PREDICT</strong>.</li></ul><h3>Step 6 — Enrich and Integrate</h3><p>Augment internal data with <strong>external sources</strong>, geographic, demographic, or market data, to add context and improve model accuracy, creating a more comprehensive dataset.</p><p>Enrich and enhance existing data by <strong>adding columns that pre-determine AI values </strong>on columns that have static meanings. For example, update your ML_ tables with AISQL in Snowflake and add a Category column to classify a product review that was purchased. This will reduce the expenses associated with data generation, making iterative reasoning more feasible and practical for real-world applications.</p><ul><li>Snowflake AISQL Example:</li></ul><pre>SELECT AI_CLASSIFY(product_review,<br>   ARRAY_CONSTRUCT(&#39;Positive&#39;, &#39;Negative&#39;, &#39;Neutral&#39;)) <br>     AS sentiment_category</pre><h3>Step 7 — Structure and Label for Learning</h3><p>Organize data logically and <strong>label it appropriately</strong> for supervised learning. Perform <strong>feature engineering</strong>, like extracting “time of day” from timestamps, to enhance predictive power.</p><p>Design data labeling by defining clear categories, consistent rules, and quality control. For example, in a customer support chatbot project, label intents like “refund request,” “product inquiry,” or “technical issue” using a detailed guide with annotated examples. Train labelers on edge cases, use multiple annotators per sample, and resolve discrepancies through consensus to ensure accuracy and consistency.</p><h3>Step 8 — Ensure Security and Compliance</h3><p>Protect sensitive information by <strong>anonymizing PII</strong>, implementing access controls, and adhering to regulations like <strong>GDPR or HIPAA</strong>. Trust and compliance are essential for ethical AI.</p><h3>Step 9 — Automate Curated Data with Pipelines</h3><p>Build <strong>automated data pipelines</strong> for ingestion, transformation, and storage into a curated Intelligence Layer using scalable solutions like <strong>Snowflake</strong> to handle diverse data types and support real-time processing needs.</p><p>Curating a copy of relational data into flattened input tables simplifies complex, nested structures into a more straightforward, tabular format that is easier for models to interpret accurately.</p><p>This flattened format allows for more reliable and consistent data iteration, enabling users to refine queries and focus on specific results without the model being misled by complex or ambiguous hierarchical relationships.</p><ul><li>Tools like pandas in Python support this process through functions like json_normalize, which can flatten nested JSON data.</li></ul><h3>Step 10 — Store Development, Training, Validation, Testing and Output Datasets in Intelligence Layer</h3><p>The intelligence layer relies on training, validation and testing datasets to ensure robust, accurate, and trustworthy model development and deployment.</p><h3>Development Datasets</h3><p>Storing development datasets that consist of only a few records to be used during development helps reduce compute, as well as significantly reduce development time, so that developers can <strong>perform</strong> <strong>many variations</strong> <strong>while working out necessary syntax</strong>.</p><p>Through a trial-and error process, developers may find they want to de-normalize or bring data into the Intelligence Layer in a different layout to make things more effective for queries to perform reasonably.</p><h3>Training Datasets</h3><p>A training dataset is one that is used to build a machine learning model and is used to fit the model by adjusting its internal parameters to recognize patterns in the data.</p><ul><li>For example, in a classification task to identify pets as cats or dogs based on weight and fluffiness, the training set teaches the model the relationship between these features and the correct labels.</li></ul><h3>Validation Datasets</h3><p>A validation dataset is one used to compare multiple models built on the same training dataset with hyperparameters <strong>to fine-tune and detect overfitting </strong>during model development.</p><ul><li>For instance, when training a neural network, the validation set helps detect overfitting by monitoring performance, if the model’s accuracy on the validation set starts to drop while training accuracy improves, it indicates overfitting, prompting adjustments like early stopping or regularization.</li></ul><h3>Training Datasets</h3><p>A testing dataset is one used to evaluate the final model on unseen data to assess real-world performance.</p><p>This ensures the model <strong>generalizes well beyond the training data and reflects its true effectiveness in practical applications</strong>. Without a separate test set, the model’s performance metrics could be overly optimistic, leading to unreliable predictions in production environments.</p><h3>Output Datasets</h3><p>In addition, storing output datasets in the Intelligence layer, provides the ability to <strong>iterate on results and use them as input </strong>for next iteration evaluation to tune intelligence even further.</p><ul><li>Adding a <strong>max_tokens</strong> parameter within natural language prompts helps the output dataset maintain effectivity, as well as stay within reasonable cost. This parameter <strong>controls the maximum number of tokens</strong> in the generated output, effectively limiting the size of the response.</li><li>Snowflake AISQL Example:</li></ul><pre>SELECT AI_COMPLETE( &#39;WITH country_sales AS (<br>    SELECT country<br>           , SUM(amount) AS total_sales <br>    FROM sales GROUP BY country <br>) <br>SELECT country_sales.country<br>       , total_sales <br>FROM country_sales <br>WHERE &#39;, model =&gt; &#39;gpt-4-turbo&#39;, temperature =&gt; 0.7, max_tokens =&gt; 50 ) <br>    AS completion;</pre><ul><li>Additionally, when using the <strong>snowflake.cortex.complete </strong>function, you can structure your prompt to explicitly <strong>instruct the model to limit the output length</strong>, such as by specifying a word or token count. For instance, a prompt can be crafted to generate a summary in less than 200 words, which inherently restricts the output size.</li><li>Snowflake AISQL Example:</li></ul><pre>SELECT <br> SNOWFLAKE.CORTEX.COMPLETE(&#39;snowflake-arctic&#39;<br>        , &#39;Summarize Q2 performance in less than 200 words.&#39;) <br>   AS summary;</pre><h3>Step 11 — Validate, Monitor and Certify Continuously</h3><p>Implement validation, monitoring and certification rules to <strong>maintain data quality over time</strong>. Regular audits and feedback loops ensure your data, and AI models, evolve effectively.</p><p>Implementing a lightweight <strong>AI Ready Data Certification framework</strong> metadata engine that applies a Readiness Score to your data applies ongoing monitoring of your Intelligence Layer to alert you of anomalies, drift, insufficiencies, reduction of key attribute conformance and other conditions that are otherwise unseen that can reduce the effectiveness of information.</p><h3>Conclusion</h3><p>AI-ready data is the cornerstone of successful artificial intelligence initiatives, transforming raw information into a powerful asset that drives innovation, accuracy, and trust in AI systems. As organizations increasingly rely on AI to make critical decisions, the importance of clean, structured, and ethically managed data becomes essential. By prioritizing data quality, timeliness, and governance, businesses can unlock the full potential of AI, ensuring models are not only intelligent but also reliable and fair.</p><p>The journey to AI readiness is not a one-time effort but an ongoing commitment to data excellence, because in the world of AI, the quality of your information depends entirely on its readiness and can produce data products that are <strong>a direct asset of your organization</strong>.</p><p>The journey to AI readiness is not a one-time effort but an ongoing commitment to data excellence, because in the world of AI, the quality of your information depends entirely on its readiness and can produce data products that are a direct asset of your organization.</p><p><em>Related article: AI and The Intelligence Layer</em></p><p><a href="https://medium.com/@angelamarieharney/ai-and-the-intelligence-layer-b3458cfdea77"><em>https://medium.com/@angelamarieharney/ai-and-the-intelligence-layer-b3458cfdea77</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=346601c73b20" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI and The Intelligence Layer]]></title>
            <link>https://medium.com/@angelamarieharney/ai-and-the-intelligence-layer-b3458cfdea77?source=rss-7e068c9737ea------2</link>
            <guid isPermaLink="false">https://medium.com/p/b3458cfdea77</guid>
            <dc:creator><![CDATA[Angela Harney]]></dc:creator>
            <pubDate>Mon, 15 Dec 2025 00:08:12 GMT</pubDate>
            <atom:updated>2025-12-15T01:10:37.131Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/812/1*3vWuUOHpwrm4xv92HEvjLw.png" /></figure><p>In modern times, the most successful organizations aren’t just using AI — they’re building an intelligence layer that orchestrates their entire operation. This isn’t about deploying another tool that writes emails or scrapes data. It’s about creating a central nervous system for your business, where AI connects every system, learns from every interaction, and empowers human decision-making. It’s a Universal Dictionary. The difference between a company that’s merely adopting AI and one that’s truly transforming its operations comes down to this — are you using AI as a hammer, or as an operating system?</p><p><strong>The Problem with the Tool Mentality</strong></p><p>Most founders start with the wrong mindset. They find a tool that promises “AI-powered email generation” and plug it in, expecting instant results. The outcome? Low open rates, poor reply rates, and even domain blacklisting. The conclusion? “AI doesn’t work.” But that’s like blaming a hammer because you tried to drive a screw with it. The issue isn’t the tool — it’s the approach.</p><p>When AI is used as a standalone tool, it’s limited to one task — generating a message, finding a lead, or analyzing a dataset. It doesn’t learn from your CRM, your sales pipeline, or your marketing campaigns. It operates in isolation, producing generic, context-free outputs that fail to resonate with real people.</p><p><strong>What Is an Intelligence Layer?</strong></p><p>An intelligence layer is not a tool. It’s the connective tissue that binds your entire tech stack into a single, intelligent system. It pulls data from every source — CRM, email platforms, LinkedIn, website behavior, support tickets — and uses that data to make real-time, context-aware decisions. It learns from every interaction, adapts to new signals, and surfaces insights that humans would never see on their own.</p><p>Imagine this — instead of an SDR manually copying a prospect’s info from LinkedIn into a CRM, the intelligence layer monitors thousands of profiles overnight. It detects a funding announcement, a new hire, or a competitor mention. It cross-references that signal with your ideal customer profile (ICP), enriches the data with firmographic details from six sources, and generates a personalized message based on industry trends, past success patterns, and current pipeline context. The human only needs to approve or tweak the message before it’s sent. One decision. The rest is automated intelligence.</p><p><strong>The Three Layers of Intelligent GTM</strong></p><p>Every effective intelligence layer operates on three interconnected layers:</p><ol><li><strong>Data Collection — </strong>This is the foundation. The layer gathers data from CRM, email engagement, website behavior, LinkedIn interactions, and support trends. Without this, AI has no raw material to work with.</li><li><strong>Intelligence (AI) — </strong>This is where AI earns its keep. It identifies hidden patterns, predicts which leads will convert, suggests the optimal message for each individual, and continuously learns from every outcome. It’s not just generating content — it’s making strategic decisions based on real-time context.</li><li><strong>Human Strategy — </strong>This is the crucial final piece. AI suggests. Humans decide. The intelligence layer surfaces insights, but your team makes the strategic calls — “Should we pivot messaging for this vertical?” “Is this lead worth a human touchpoint?” “How do we turn this pattern into a repeatable system?”</li></ol><p>This model isn’t theoretical. Clients using this intelligence layer have seen reply rates jump from 2% to 8%, demos increase from 10 to 40 per month, and pipelines shift from random to predictable. The same team. The same effort. A different operating system.</p><p><strong>Powering Intelligence with a Metadata Engine</strong></p><p>At the heart of a true intelligence layer lies the metadata engine — a dynamic system that doesn’t just collect data, but interprets, connects, and activates it across the go-to-market stack. Unlike static integrations, a metadata engine continuously normalizes data from CRM, MAP, CDPs, and ad platforms, creating a unified, real-time view of the buyer journey.</p><p>It powers AI agents that autonomously run thousands of micro-experiments, adjusting audiences, creatives, and bids based on downstream pipeline impact, not just surface-level engagement.</p><p>By optimizing for revenue signals — like opportunities and closed deals — and filtering out bot-like noise, the engine ensures marketing efforts align with business outcomes.</p><p>Successful platforms exemplify this, using agentic AI to unify intent data (e.g., buyer activity), automate campaign execution, and enable teams to act with the precision of an F1 pit crew — turning fragmented workflows into a coordinated, self-optimizing growth engine.</p><p><strong>Designing Your Intelligence Layer — Key Principles</strong></p><p>Building an intelligence layer isn’t just about choosing the right AI tool. It’s about system design. Here are the core principles:</p><ul><li><strong>Connectivity is King — </strong>The layer must integrate with every system in your stack — CRM, marketing automation, email, analytics — so data flows freely and decisions are informed by the full picture.</li><li><strong>Scalability and Performance — </strong>Design modularly. Keep ingestion, storage, modeling, and visualization layers decoupled so components can scale independently. Use materialized views, partitioning, and caching to ensure dashboards load in seconds, not minutes.</li><li><strong>Human-Centric Design — </strong>The AI should empower people, not replace them. Like the Accenture Logistics Platform, which advises postal workers but allows them to override suggestions, the system must maintain human autonomy while learning from real-world feedback. Transparency is key — users should understand why a recommendation was made, even if it’s just a simple “Recommended because you liked this”.</li><li><strong>Iterative Development — </strong>Start with a clear problem. Use AI to research real pain points from forums and social media. Break the problem into stages, develop hypotheses, and test them with data. Use tools like Perplexity or ChatGPT with Search to generate research-backed content and refine your ideas.</li></ul><p><strong>The Future Is Intelligent</strong></p><p>The founders winning in modern times aren’t the ones with the best AI tools. They’re the ones who built an intelligence layer that orchestrates everything. They’ve moved beyond the “hammer” approach and embraced AI as an operating system — a system that learns, adapts, and scales with their business. It turns their data into a company asset.</p><p>If you’re still using AI as a tool, it’s time to upgrade. Ask yourself — “Is my AI a tool, or an intelligence layer?” The answer will determine whether you’re just keeping up — or leading the future.</p><p><em>Related Article: AI-Ready Data: The First Step in the AI Journey to unlocking Intelligent Insights</em></p><p><a href="https://medium.com/@angelamarieharney/ai-ready-data-the-first-step-in-the-ai-journey-to-unlocking-intelligent-insights-346601c73b20"><em>https://medium.com/@angelamarieharney/ai-ready-data-the-first-step-in-the-ai-journey-to-unlocking-intelligent-insights-346601c73b20</em></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b3458cfdea77" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Using Cortex AI to Predict what Customers will Pay in the Future]]></title>
            <link>https://medium.com/@angelamarieharney/using-cortex-ai-to-predict-what-customers-will-pay-in-the-future-83e4f94fc17b?source=rss-7e068c9737ea------2</link>
            <guid isPermaLink="false">https://medium.com/p/83e4f94fc17b</guid>
            <dc:creator><![CDATA[Angela Harney]]></dc:creator>
            <pubDate>Fri, 28 Nov 2025 06:15:33 GMT</pubDate>
            <atom:updated>2025-11-28T23:54:22.527Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/822/1*AoFR17xkicluWUTmp4U6lg.png" /></figure><p>My idea was to create a solution that recommends prices for products. The solution would answer the question of what price customers would pay in the future based on customer reviews. Would customers pay more for the product or want to start paying less?</p><p>An additional challenge was to also recommend other products each customer might buy based on certain factors.</p><p>A final challenge was to have the solution recommend marketing campaigns for the additional product recommendations.</p><h3>Step #1 — Create AI Ready Data</h3><p>Create a product purchases and customer sentiment table.</p><ul><li>What I learned here was that creating a table that combined purchase data and customer sentiment helped me to arrive at what I was after with a more simplistic approach.</li></ul><blockquote>Tip: Create a layer of AI ready data close to the business layer that combines data with conditional information.</blockquote><h3><strong>Step #2 — Classify the Data</strong></h3><p>Use AI_CLASSIFY to classify data into user-defined categories.</p><ul><li>This will help identify what matters to you. How positive does something have to be in order to be meaningful. What kind of statements are negative and what do you consider to be neutral.</li></ul><blockquote>Tip: Using Snowflake’s built-in functions for this makes this step quick so that you can focus on the core outcomes.</blockquote><h3>Step #3 — Use the classified data to Predict prices</h3><p>Use AI_AGG to take in a natural language prompt to ask for outcomes based on conditions</p><ul><li>What I found was that keeping the prompt simple allowed a lot more information to be returned than I could have thought to ask for.</li></ul><blockquote>Tip: Learning basics on Prompt Engineering will help you here, but you don’t have to be too advanced to get good initial results</blockquote><h3>Step #4 — Preview customer sentiment</h3><p>Use a natural language prompt with Snowflake’s CORTEX AI_FILTER function to sift through customer reviews to get a list of products that customers consider to be overpriced.</p><ul><li>Prior to asking for product recommendations, I wanted to see what customers thought of certain pricing aspects.</li></ul><blockquote>Tip: AI filtering can apply expressions to data that can otherwise be too complex to achieve with direct query syntax.</blockquote><h3>Step #5 — Use product purchases and reviews to get Product recommendations</h3><p>Use Snowflake’s AI_AGG built-in function again to get recommendations of products with an altered natural language prompt and based on different factors.</p><ul><li>It was amazing to see the specific products that were recommended along with the reason why they were recommended.</li></ul><blockquote>Tip: Snowflake’s LLMs return product information from curated sources along with identifiable reasoning.</blockquote><h3>Step #6 — Suggest Marketing Campaigns for Product recommendations</h3><p>Use the AI_AGG built-in function again to request that marketing campaigns be recommended for each product.</p><ul><li>I learned that with the natural language prompt you can tell it to use a specific aspect of a customer review, such as product appreciation so that you can focus the marketing on why customers will enjoy the products.</li></ul><blockquote>Tip: Snowflake’s LLMs return detailed and actionable information that is creative as well as reasonable.</blockquote><h3>Step #7 — Create a Price Prediction Data Pipeline</h3><p>Create a Snowflake Stream and a Task that contains a query utilizing Cortex AI_COMPLETE that leverages the “snowflake-arctic” LLM.</p><ul><li>The task can be set to run weekly and insert a new set of rows into an ongoing price prediction table.</li></ul><blockquote>Tip: You can use dynamic SQL to combine columns from the customer purchases table into the natural language prompt.</blockquote><h3>Step #8 — AI Pricing and Speed of Development</h3><p>Code for this solution can be found in the Appendix of this article. It cost $2.77 in a trial account to try out results, tweak the natural language prompts and confirm the table design. The solution was accomplished in 1 hour.</p><ul><li>Snowflake charges per million input and output tokens based on what LLM you utilize. Snowflake is not currently charging for Snowflake Copilot, but may elect to do so in future. The secret to success with regards to any AI cost is to limit the number of records you use as input. The gold rush of the future will be to design AI Ready Data that is the most effective it can be for the smallest size.</li></ul><blockquote>Tip: AISQL in Snowflake is now the standard upon which future functionality will be built.</blockquote><p>Enjoy the road to AI with Cortex AI and AISQL. Snowflake is also carving the way with Snowflake MCP Servers. It will be amazing to watch how the landscape changes in the upcoming years.</p><blockquote>References:</blockquote><blockquote><a href="https://docs.snowflake.com/en/user-guide/snowflake-cortex/aisql">Snowflake Cortex AI Functions (including LLM functions) | Snowflake Documentation</a></blockquote><blockquote><a href="https://select.dev/posts/snowflake-cortex-ai-sql-overview-and-cost-monitoring">Snowflake Cortex AI SQL: Overview, Pricing &amp; Cost Monitoring</a></blockquote><h3>APPENDIX</h3><p>The following code can be executed in a trial account. It creates a small database with a few tables that contain less than 100 rows per table. It creates a Stream and a Task as a data pipeline. It generates a SQL scripting stored procedure to generate a custom number of sample rows. To control costs, suspend or drop the task when done reviewing the solution.</p><blockquote>use role accountadmin;<br>use warehouse compute_wh;<br>create database if not exists stayfitdata;<br>use database stayfitdata;<br>create schema if not exists analytics;<br>use schema analytics;</blockquote><blockquote>— CREATE THE TABLE</blockquote><blockquote>CREATE OR REPLACE TABLE customer_bike_purchases (<br> bike_product_id VARCHAR,<br> bike_terrain_type VARCHAR,<br> purchase_city VARCHAR,<br> product_review VARCHAR,<br> purchase_price NUMBER(10, 2),<br> purchase_date date<br>);</blockquote><blockquote>— select * from customer_bike_purchases;</blockquote><blockquote>— PROCEDURE THAT GENERATES A SET NUMBER OF N ROWS OF DATA</blockquote><blockquote>CREATE OR REPLACE PROCEDURE generate_bike_purchase_data(target_row_count INT)<br>RETURNS STRING<br>LANGUAGE SQL<br>AS<br>$$<br>DECLARE<br> bike_ids ARRAY := ARRAY_CONSTRUCT(‘B001’, ‘B002’, ‘B003’);<br> terrains ARRAY := ARRAY_CONSTRUCT(‘Mountain’, ‘Road’, ‘Hybrid’);<br> cities ARRAY := ARRAY_CONSTRUCT(‘Denver’, ‘Boulder’, ‘Seattle’);<br> reviews ARRAY := ARRAY_CONSTRUCT(‘Great bike!’, ‘Good value’, ‘Needs improvement’, ‘Excellent ride’);<br> row_count INT := 0;<br> bike_id VARCHAR;<br> terrain_type VARCHAR;<br> city_name VARCHAR;<br> review_text VARCHAR;<br> price INT;<br> purchase_dt DATE;<br>BEGIN<br> FOR i IN 0 TO ARRAY_SIZE(:bike_ids) — 1 DO<br> bike_id := GET(:bike_ids, :i);<br> FOR j IN 0 TO ARRAY_SIZE(:terrains) — 1 DO<br> terrain_type := GET(:terrains, :j);<br> FOR k IN 0 TO ARRAY_SIZE(:cities) — 1 DO<br> city_name := GET(:cities, :k);<br> review_text := GET(:reviews, MOD(:i + :j + :k, 4));<br> price := 500 + (RANDOM() % 500)::INT;<br> <br> IF (row_count &gt;= :target_row_count) THEN<br> RETURN ‘Target row count reached.’;<br> END IF;</blockquote><blockquote>purchase_dt := DATEADD(‘day’, UNIFORM(0, 1095, RANDOM()), ‘2023–01–01’::DATE);</blockquote><blockquote>INSERT INTO customer_bike_purchases <br> (bike_product_id, bike_terrain_type, purchase_city, product_review, purchase_price, purchase_date)<br> VALUES (:bike_id, :terrain_type, :city_name, :review_text, :price, :purchase_dt);</blockquote><blockquote>row_count := row_count + 1;<br> END FOR;<br> END FOR;<br> END FOR;<br> RETURN ‘Generated ‘ || row_count || ‘ rows.’;<br>END;<br>$$;</blockquote><blockquote>— CALL SPROC</blockquote><blockquote>— this will generate a random set so many insert less actual rows<br>CALL generate_bike_purchase_data(80);</blockquote><blockquote>select * from customer_bike_purchases;</blockquote><blockquote>— ANALYZE CUSTOMER SENTIMENT — AI_CLASSIFY</blockquote><blockquote>CREATE OR REPLACE TRANSIENT TABLE aiclassify_customer_sentiment as<br>SELECT <br> bike_product_id,<br> bike_terrain_type,<br> purchase_city,<br> product_review,<br> purchase_price,<br> SNOWFLAKE.CORTEX.SENTIMENT(product_review) AS review_sentiment,<br> AI_CLASSIFY(product_review, ARRAY_CONSTRUCT(‘Positive’, ‘Negative’, ‘Neutral’)) AS sentiment_category<br>FROM customer_bike_purchases;</blockquote><blockquote>select * from aiclassify_customer_sentiment;</blockquote><blockquote>— PREDICT WILLINGNESS TO PAY HIGHER OR LOWER — AI_AGG</blockquote><blockquote>CREATE OR REPLACE TRANSIENT TABLE aiclassify_predict_prices as<br>SELECT <br> bike_product_id,<br> AI_AGG(<br> product_review,<br> ‘Summarize customer feedback on value for money and suggest a price range that reflects customer sentiment.’<br> ) AS value_perception_summary<br>FROM customer_bike_purchases<br>GROUP BY bike_product_id;</blockquote><blockquote>select * from aiclassify_predict_prices;</blockquote><blockquote>— IDENTIFY REVIEWS THAT MENTION SPECIFIC FEATURE — AI_FILTER</blockquote><blockquote>— if querying a table</blockquote><blockquote>CREATE OR REPLACE TRANSIENT TABLE aifilter_specific_feature as<br>SELECT <br> bike_product_id,<br> purchase_price,<br> product_review<br>FROM customer_bike_purchases<br>WHERE SNOWFLAKE.CORTEX.AI_FILTER(<br> PROMPT(‘The customer thinks the bike is too expensive: {0}’, product_review)<br>) = TRUE;</blockquote><blockquote>select * from aifilter_specific_feature;</blockquote><blockquote>— PREDICT OTHER BIKE PURCHASES — AI_AGG</blockquote><blockquote>CREATE OR REPLACE TRANSIENT TABLE aiagg_predict_purchases as<br>SELECT <br> bike_product_id,<br> AI_AGG(<br> product_review,<br> ‘Summarize customer feedback on product appreciation and suggest what other bike the customer would buy based on customer sentiment.’<br> ) AS value_perception_summary<br>FROM customer_bike_purchases<br>GROUP BY bike_product_id;</blockquote><blockquote>select * from aiagg_predict_purchases;</blockquote><blockquote>— PREDICT MARKETING CAMPAIGNS — AI_AGG</blockquote><blockquote>CREATE OR REPLACE TRANSIENT TABLE aiagg_marketing_campaigns as<br>SELECT <br> bike_product_id,<br> AI_AGG(<br> product_review,<br> ‘Summarize what marketing campaign utilize customer feedback on product appreciation and suggest what other bike the customer would buy.’<br> ) AS value_perception_summary<br>FROM customer_bike_purchases<br>GROUP BY bike_product_id;</blockquote><blockquote>select * from aiagg_marketing_campaigns;</blockquote><blockquote>— CREATE ML PIPELINE</blockquote><blockquote>— UPDATE PREDICTIONS WHEN NEW DATA ARRIVES</blockquote><blockquote>CREATE OR REPLACE STREAM bike_purchase_stream<br>ON TABLE customer_bike_purchases<br>APPEND_ONLY = TRUE<br>SHOW_INITIAL_ROWS = TRUE;</blockquote><blockquote>CREATE OR REPLACE TRANSIENT TABLE predicted_prices (<br> bike_product_id VARCHAR, <br> predicted_price VARCHAR, <br> city VARCHAR, <br> updated_at date<br>);</blockquote><blockquote>— CREATE A TASK TO PROCESS THE DATA</blockquote><blockquote>CREATE OR REPLACE TASK update_price_predictions<br>WAREHOUSE = COMPUTE_WH — Replace with your warehouse name<br>SCHEDULE = ‘7 DAYS’<br>WHEN SYSTEM$STREAM_HAS_DATA(‘bike_purchase_stream’)<br>AS<br>BEGIN<br> — Example: Insert predictions into a target table using AI functions<br> INSERT INTO predicted_prices (bike_product_id, predicted_price, city, updated_at)<br> SELECT <br> bike_product_id,<br> AI_COMPLETE(‘snowflake-arctic’, <br> ‘Predict future price for ‘ || bike_product_id || ‘ in ‘ || purchase_city || <br> ‘ based on terrain: ‘ || bike_terrain_type || ‘ and review: ‘ || product_review<br> ) AS predicted_price,<br> purchase_city,<br> CURRENT_TIMESTAMP()<br> FROM bike_purchase_stream;<br>END;</blockquote><blockquote>— START THE TASK</blockquote><blockquote>ALTER TASK update_price_predictions RESUME;</blockquote><blockquote>EXECUTE TASK update_price_predictions;</blockquote><blockquote>select * from predicted_prices;</blockquote><blockquote>— Be sure to suspend the task when done with development<br>ALTER TASK update_price_predictions SUSPEND;</blockquote><blockquote>—</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=83e4f94fc17b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Power Duo — Snowflake and DataOps.live — Part 5 of 5: Hands-On Lab Overview]]></title>
            <link>https://medium.com/@angelamarieharney/the-power-duo-snowflake-and-dataops-live-part-5-of-5-hands-on-lab-overview-80d10c9bb1dd?source=rss-7e068c9737ea------2</link>
            <guid isPermaLink="false">https://medium.com/p/80d10c9bb1dd</guid>
            <dc:creator><![CDATA[Angela Harney]]></dc:creator>
            <pubDate>Sun, 24 Aug 2025 19:37:36 GMT</pubDate>
            <atom:updated>2025-08-24T19:37:36.724Z</atom:updated>
            <content:encoded><![CDATA[<h3>The Power Duo — Snowflake and DataOps.live — Part 5 of 5: Hands-On Lab Overview</h3><p>This article covers the steps you will be doing when you particpate in a hands-on lab with DataOpslive’s new native apps for CI/CD. Dynamic Delivery and Dynamic Transformation are two apps that you can use for integrated CI/CD Snowflake and you will love the integration with Git and the VSCode-like experience that DataOps.live provides for development tools in the cloud.</p><p>This article provides an overview of their hands-on labs so that you can get the most out of it when you sign-up for your session.</p><h3>Upcoming Hands-on Lab Sessions →</h3><p>Visit the DataOps.live web-site for ongoing details about upcoming hands-on lab sessions, including this one on AUG 28, 2025.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/716/1*1RI2Zcs27R_OVjxgZsiHcQ.png" /></figure><blockquote><a href="https://us06web.zoom.us/webinar/register/WN_Y1tc8qIERoaSwrSMJkFavw#/registration">Webinar Registration — Zoom</a> — <strong>August 28, 2025</strong></blockquote><h3><strong>Win Free Access to DataOps.live Platform →</strong></h3><p>Don’t miss your chance to enter a raffle to win free access to the DataOps.live platform. This cloud platform tool compliments Snowflake and makes it easy to deploy and run jobs to integrate seamlessly.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/476/1*abLYZky5vkr1Q3Yt2vU5Cw.png" /><figcaption><a href="https://www.dataops.live/enter-the-raffle-get-your-free-apps">Enter the raffle, get your free apps</a></figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Mlk0mva3b4paNs4zuGEhcg.png" /></figure><h3><strong>Hands-On Lab Overview</strong></h3><p>The following screen shots are intended to give you a high-level look at the DataOps.live platform apps called Dynamic Delivery and Dynamic Transformations so you can see how easy it is to interact with these tools.</p><h3><strong>Get DataOps.live Tools from Marketplace</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*CqRiLUSeTd2JulX3IEjwXQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IOnTWnU_a--jYYhc0n3HcA.png" /></figure><h4><strong>Select a Dyanamic Tool in DataOps.live Platform</strong></h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*cMbRT5F4HYHUkI1o1QlmLg.png" /></figure><h3><strong>Dynamic Delivery Setup</strong></h3><p>Once inside the DataOps.live platform, when you select Dynamic Delivery, it will take you through several steps to setup the tool and configure what database and schema you would like to setup delivery for.</p><p>There are some basic privileges that will be applied in the process.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/914/1*hUM-BBZoXN34u13qtJDezQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/658/1*sHSTpa9fzRbMJyN8HOCK8A.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Hg6qNb0SwXoesX4FyFVH8A.png" /></figure><h4>Git Project Integration</h4><p>Once setup is complete, you will be taken to the git project where you can interact with files that have been setup for you of the database and schema objects you selected.</p><p>Notice that Jinja reference and environment tags were used in the .yml files that were created.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*YP2H23GSK_uwf-rXJj24pg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LJQlnfVDnvp8zXJPSWDfBQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_fgEgqb_dauEtPA0V0V1eA.png" /></figure><h4>Git Code Merge Requests</h4><p>Click Merge Requests at the left in the DataOps.live platform and you will see the last branch check-in that was automatically saved.</p><p>Once you click Create Merge Request, you will be taken through a set of steps to provide a Title, Assignee and Reviewer.</p><p>There is a panel at the right where the Reviewer and Assignee can interact with the Merge Request, including view file differences.</p><p>The Merge can be done with standard options available, such as Delete Source Branch when done.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HeXJAEVS4gasvKnO6xaICw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MLjNmBo-fzSZ8GDSzmPhCw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mgyXf1xQYOikAbMUXk3cUA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Q4CaBoZlkVAAcuSeUp4WBw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hBd5SpOq2KQ_-9sDq7BYyQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ubNobsr2T1qFn810oYl2-g.png" /></figure><h4>Runing A Pipeline</h4><p>Once a branch has been established, you can run it as a pipeline. DataOps.live provides a panel for viewing the steps of the pipeline while they are executing so you can easily manage progress.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ai8yVMmlNvGD7pqe81WIxA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IK2B365tOID06vaXM0vhfQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LvPx7oMUIG7xpw6pBfTIOg.png" /></figure><h4>Executing a Job</h4><p>After a pipeline has been ran, you can execute a job to Setup Snowflake and create the objects you established in a DEV or TST location.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*H1k_MKuU2BfBbDHozyWyZw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LvPx7oMUIG7xpw6pBfTIOg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-aQ7WQEXpaxcRcTuP7igrQ.png" /></figure><h3>Dynamic Transformations Setup</h3><p>From within the DataOps.live platform, you can also run transformations using the Dynamic Transformations app. You will be asked to identify a Source and Target dbt project, as well as a Target Snowflake Database.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/961/1*mEhcRgbH97CCcg2sWmSH7A.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PmrQqcyoG-jT0DregQwNcg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/675/1*GZqsh4vY7AUp4FR9er4tSQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/683/1*hc7Z3_D0vwaHtnnGSBEj9A.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/733/1*-nJmxZ1dxPCbdmDRRLUsDQ.png" /></figure><h4>Develop the Transformations</h4><p>Once you have completed the setup of Dynamic Transformations, you will be taken to the main repositor view where you can click Develop.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nfcqwmVwHeu6iKg1QbyNWQ.png" /></figure><p>This will create and startup an application experience similar to VSCode in the cloud within the DataOps.live platform.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/762/1*IFHTUJGFjEivsQ714sbX2A.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hqud-KtsIScEl_8_v8y-vA.png" /></figure><p>Once it is setup, you can begin configuring the environment in a Gitpod. Gitpod is a cloud-based development environment (CDE) platform that provides on-demand, pre-configured development environments for software development. It aims to streamline the development workflow by eliminating the need for developers to set up and maintain local development environments.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*HjcpDpWT3dPCDZDxdtTmZA.png" /></figure><p>It comes complete with a Unix-based terminal for issuing Git and other execution commands.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2EtbzVcvA-L_cVnQNKJPnA.png" /></figure><h4>DataOps.live Copilot</h4><p>The DataOps.live Assistant provides outstanding Copilt assistance to code .sql file, models and much more to simplify the creation of your solution components.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8KKwdnjkc7PkocEoehu7xg.png" /></figure><h4>Execute SQL from a File</h4><p>You can execute SQL in a file and see the resultset from with DataOps.live.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IlDP9i6_8ZnSv_xRBruvFQ.png" /></figure><h4>Create Models</h4><p>You can generate and execute models.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-g9cy10OHxD7z4vIFmgaVg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1003/1*zm46U_oJZAXTUUrcU1EHdg.png" /></figure><h4>Run Transformation Pipeline</h4><p>After running a pipeline to generate necessary objects in Snowflake, you can then build the models that were created.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*-wixo9Dy0hcza0R-FRTplQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/869/1*lfJL5IXHuEoYzWN_rWjMqA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*s-1Po0R92ElnbRBvJS2IMw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*o63HLZeDKnqqB7c2-IpvnA.png" /></figure><h4>Generate Documentation</h4><p>DataOps.live also provides a documentation experience that allows you to search a catalog of objects to see details about a table or view and see a graphical lineage depicting relationships as well.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*t-amhAPaFrBPpoUbewE4wg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Q8Ob9p3bkQpBuj1bvRkYDw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*MAAqHS78IJVc2nB7g85WBA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KoQlHw10G1NIONZUTdHgzA.png" /></figure><h3>Conclusion</h3><p>Whether you have already created projects, or are just getting started, and don’t want to have to have to leave the Snowflake platform to utilize third-party tools for easy and seamless CI/CD and transformations, don’t miss taking a look at DataOps.live and getting hands-on experience with these tools, because they make it easy → and <strong><em>Snowflake will pay 500 minutes every month towards the use of these tools.</em></strong></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=80d10c9bb1dd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Power Duo — Snowflake and DataOps.live — Part 4 of 5: Data Workflows Git Integration]]></title>
            <link>https://medium.com/@angelamarieharney/the-power-duo-snowflake-and-dataops-live-part-4-of-5-data-workflows-git-integration-db7079d1dcff?source=rss-7e068c9737ea------2</link>
            <guid isPermaLink="false">https://medium.com/p/db7079d1dcff</guid>
            <dc:creator><![CDATA[Angela Harney]]></dc:creator>
            <pubDate>Sun, 24 Aug 2025 19:37:30 GMT</pubDate>
            <atom:updated>2025-08-24T19:37:30.169Z</atom:updated>
            <content:encoded><![CDATA[<h3>The Power Duo — Snowflake and DataOps.live — Part 4 of 5: Data Workflows Git Integration</h3><p>In today’s data-driven landscape, the ability to rapidly develop, deploy, and manage data solutions is crucial. Modern platforms like Snowflake offer the scalability and flexibility to handle vast amounts of data, while practices like DevOps and DataOps provide the frameworks to streamline development and operations. Combining these forces, the integration of Git with Snowflake and platforms like DataOps.live is revolutionizing data workflows, bringing increased efficiency, collaboration, and reliability to the data pipeline.</p><h3><strong>The Power of Snowflake’s Git Integration →</strong></h3><p>Snowflake’s native Git integration, now generally available on all clouds, allows organizations to treat their data environments as code, ushering in a new era of DataOps for the platform. By connecting Snowflake to external Git repositories (such as GitHub, GitLab, Azure DevOps, and BitBucket), developers can:</p><ul><li><strong>Version Control All Snowflake Assets:</strong> Store and manage all Snowflake objects (SQL scripts, Snowpark code, Native Apps, Streamlit Apps, etc.) under version control, enabling a single source of truth for all code and configurations.</li><li><strong>Streamline Development and Collaboration:</strong> Enable multiple developers to work concurrently on projects without conflicts, using established Git workflows like branching and merging.</li><li><strong>Automate Deployments:</strong> Use repository stages in Snowflake to directly deploy code changes to different environments (development, testing, production) automatically, removing manual steps and ensuring consistency.</li><li><strong>Rollback and Audit:</strong> Leverage Git’s version history to easily revert to previous versions of code or configurations if needed, while also maintaining a complete audit trail of all changes.</li></ul><h3><strong>DataOps.live → Orchestrating the Snowflake Data Cloud</strong></h3><p>While Snowflake’s native Git integration provides the foundation, platforms like DataOps.live elevate the DataOps experience by offering a comprehensive control plane for orchestrating the entire data lifecycle within Snowflake. DataOps.live complements Snowflake’s capabilities by:</p><ul><li><strong>Treating the Data Platform as Code:</strong> Storing configurations as code and utilizing Git for source control, enabling the building and rebuilding of the data platform from a Git repository in minutes.</li><li><strong>Facilitating Sandboxed Environments:</strong> Creating sandboxed Snowflake environments, complete with zero-copy clones of data, that correspond to different branches in Git, accelerating development and testing.</li><li><strong>Automating Testing and Deployment:</strong> Providing tools for building, testing, and deploying data pipelines and data products across various environments, ensuring high data quality and faster time to market.</li><li><strong>Enhancing Collaboration:</strong> Offering a unified developer experience that simplifies collaboration between data engineers, analysts, and business users, regardless of their technical expertise.</li><li><strong>Integrating with the Snowflake Ecosystem:</strong> Integrating seamlessly with other tools and features within the Snowflake ecosystem, such as Snowpark, Streamlit, and dbt, to offer a holistic data management solution.</li></ul><h3><strong>Conclusion →</strong></h3><p>The synergy between Snowflake’s native Git integration and DataOps.live empowers organizations to fully embrace DataOps principles, transforming how they develop, deploy, and manage data products and pipelines within the Snowflake Data Cloud. This powerful integration drives efficiency, fosters collaboration, and ultimately allows businesses to extract more value from their data, faster and with greater confidence.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=db7079d1dcff" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Power Duo — Snowflake and DataOps.live — Part 3 of 5: Dynamic Transformations]]></title>
            <link>https://medium.com/@angelamarieharney/the-power-duo-snowflake-and-dataops-live-part-3-of-5-dynamic-transformations-a46acfa76ac8?source=rss-7e068c9737ea------2</link>
            <guid isPermaLink="false">https://medium.com/p/a46acfa76ac8</guid>
            <dc:creator><![CDATA[Angela Harney]]></dc:creator>
            <pubDate>Sun, 24 Aug 2025 19:37:23 GMT</pubDate>
            <atom:updated>2025-08-24T20:07:02.077Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>The Power Duo — Snowflake and DataOps.live — Part 3 of 5: Dynamic Transformation</strong></h3><h3><strong>Transformations →</strong></h3><p>Elevating Data Transformations: <a href="https://www.dataops.live/">DataOps.live</a> Dynamic Transformation on Snowflake Marketplace</p><p>DataOps.live’s Dynamic Transformation, available as a Snowflake Native App on the Snowflake Marketplace, represents a significant advancement in operationalizing and enhancing data transformations within the Snowflake ecosystem. This offering specifically targets the challenges faced by data engineering teams in building, deploying, and managing enterprise-grade data pipelines, particularly those leveraging dbt (data build tool).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*o2kPGqBsRrkl6BKNS8hucg.png" /></figure><h3><strong>Key Capabilities and Benefits →</strong></h3><ul><li><a href="https://www.google.com/search?sca_esv=9faac44f5c63c16f&amp;rlz=1C1RXQR_enUS1077US1077&amp;cs=1&amp;q=Pipeline+Automation&amp;sa=X&amp;ved=2ahUKEwigtMnhhZaPAxVsFzQIHWNEDRgQxccNegQIHxAD&amp;mstk=AUtExfAqlCaeVAg88E9EhKpUuO49VRdCDOoS2Xa-MOUEg02MoesWK9b17gcNh31WxrtMDTe5Ja8grj77dDVVhADRO91L3aO-Wsmr4VO2seid6h0glisArR-chgugPS6FMfvmjwoUdOmiFohQmkMaIJ178c6J0Zt8fHDhIJxh3PGj6UFIsr0&amp;csui=3"><strong>Pipeline Automation:</strong></a><strong> </strong>Dynamic Transformation streamlines the entire dbt lifecycle within Snowflake. It automates critical processes including development, deployment, and environment management, reducing manual effort and accelerating time-to-value for data products. This automation extends to continuous integration and continuous delivery (CI/CD) practices for dbt projects, ensuring consistent and reliable deployments.</li><li><a href="https://www.google.com/search?sca_esv=9faac44f5c63c16f&amp;rlz=1C1RXQR_enUS1077US1077&amp;cs=1&amp;q=Enterprise-Grade+Governance+and+Control&amp;sa=X&amp;ved=2ahUKEwigtMnhhZaPAxVsFzQIHWNEDRgQxccNegQIFhAD&amp;mstk=AUtExfAqlCaeVAg88E9EhKpUuO49VRdCDOoS2Xa-MOUEg02MoesWK9b17gcNh31WxrtMDTe5Ja8grj77dDVVhADRO91L3aO-Wsmr4VO2seid6h0glisArR-chgugPS6FMfvmjwoUdOmiFohQmkMaIJ178c6J0Zt8fHDhIJxh3PGj6UFIsr0&amp;csui=3"><strong>Enterprise-Grade Governance and Control:</strong></a><strong> </strong>The solution injects robust governance capabilities into dbt projects. It enables the enforcement of data standards, facilitates compliance with organizational policies, and provides enhanced visibility across various teams and environments. This level of control is crucial for maintaining data quality and security in complex enterprise data landscapes.</li><li><a href="https://www.google.com/search?sca_esv=9faac44f5c63c16f&amp;rlz=1C1RXQR_enUS1077US1077&amp;cs=1&amp;q=Automated+Testing&amp;sa=X&amp;ved=2ahUKEwigtMnhhZaPAxVsFzQIHWNEDRgQxccNegQIGhAD&amp;mstk=AUtExfAqlCaeVAg88E9EhKpUuO49VRdCDOoS2Xa-MOUEg02MoesWK9b17gcNh31WxrtMDTe5Ja8grj77dDVVhADRO91L3aO-Wsmr4VO2seid6h0glisArR-chgugPS6FMfvmjwoUdOmiFohQmkMaIJ178c6J0Zt8fHDhIJxh3PGj6UFIsr0&amp;csui=3"><strong>Automated Testing:</strong></a><strong> </strong>Dynamic Transformation incorporates automated testing functionalities to validate data transformations. This includes both technical assertions and business quality checks, ensuring the accuracy and reliability of transformed data before it is consumed by downstream applications or analytical tools.</li><li><a href="https://www.google.com/search?sca_esv=9faac44f5c63c16f&amp;rlz=1C1RXQR_enUS1077US1077&amp;cs=1&amp;q=Seamless+Integration+with+Snowflake&amp;sa=X&amp;ved=2ahUKEwigtMnhhZaPAxVsFzQIHWNEDRgQxccNegQIHBAD&amp;mstk=AUtExfAqlCaeVAg88E9EhKpUuO49VRdCDOoS2Xa-MOUEg02MoesWK9b17gcNh31WxrtMDTe5Ja8grj77dDVVhADRO91L3aO-Wsmr4VO2seid6h0glisArR-chgugPS6FMfvmjwoUdOmiFohQmkMaIJ178c6J0Zt8fHDhIJxh3PGj6UFIsr0&amp;csui=3"><strong>Seamless Integration with Snowflake:</strong></a><strong> </strong>As a Snowflake Native App, Dynamic Transformation leverages the power and scalability of the Snowflake platform. It offers a seamless user experience, allowing data teams to manage their dbt projects and data transformations directly within their Snowflake environment. This native integration optimizes performance and simplifies data engineering workflows.</li><li><a href="https://www.google.com/search?sca_esv=9faac44f5c63c16f&amp;rlz=1C1RXQR_enUS1077US1077&amp;cs=1&amp;q=Accelerated+dbt+Project+Operationalization&amp;sa=X&amp;ved=2ahUKEwigtMnhhZaPAxVsFzQIHWNEDRgQxccNegQIHhAD&amp;mstk=AUtExfAqlCaeVAg88E9EhKpUuO49VRdCDOoS2Xa-MOUEg02MoesWK9b17gcNh31WxrtMDTe5Ja8grj77dDVVhADRO91L3aO-Wsmr4VO2seid6h0glisArR-chgugPS6FMfvmjwoUdOmiFohQmkMaIJ178c6J0Zt8fHDhIJxh3PGj6UFIsr0&amp;csui=3"><strong>Accelerated dbt Project Operationalization:</strong></a><strong> </strong>Dynamic Transformation elevates standardized dbt projects to enterprise-grade data products. By providing essential features like pipeline automation, governance, and automated testing, it enables organizations to operationalize their dbt projects with enhanced speed, reliability, and scale, turning raw data into valuable, readily consumable data products.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*r-64Qc3ZL0T5NA5H3HaEfg.png" /></figure><h3><strong>Conclusion →</strong></h3><p>In essence, Dynamic Transformation from <a href="https://www.dataops.live/">DataOps.live</a> empowers data engineering teams to achieve greater agility, efficiency, and control over their data transformation processes within Snowflake, ultimately accelerating the delivery of high-quality data products.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a46acfa76ac8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Power Duo — Snowflake and DataOps.live — Part 2 of 5: Dynamic Delivery]]></title>
            <link>https://medium.com/@angelamarieharney/the-power-duo-snowflake-and-dataops-live-part-2-of-5-dynamic-delivery-8be576f197cb?source=rss-7e068c9737ea------2</link>
            <guid isPermaLink="false">https://medium.com/p/8be576f197cb</guid>
            <dc:creator><![CDATA[Angela Harney]]></dc:creator>
            <pubDate>Sun, 24 Aug 2025 19:37:15 GMT</pubDate>
            <atom:updated>2025-08-24T19:37:15.582Z</atom:updated>
            <content:encoded><![CDATA[<h3>The Power Duo — Snowflake and DataOps.live — Part 2 of 5: Dynamic Delivery</h3><h3><strong>Dynamic Delivery from → Revolutionizing CI/CD for Snowflake</strong></h3><p>DataOps.live’s Dynamic Delivery, a Snowflake Native App available on the Snowflake Marketplace, addresses a critical need for organizations leveraging Snowflake: streamlined and automated Continuous Integration/Continuous Delivery (CI/CD) for Snowflake objects. This solution transforms the often-manual and error-prone process of managing Snowflake infrastructure into an agile, repeatable, and governed workflow.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*udxTDQ_wWh4oMy6O5YCTKA.png" /></figure><h3><strong>The Challenge →</strong></h3><p>Traditional Snowflake object management frequently involves manual SQL script deployments, inconsistent environments, and a lack of clear CI/CD frameworks. This leads to slow data delivery, increased risk of errors, and frustration for data engineering teams. Now there is a native app available on Snowflake Martketplace that is a seamless solution.</p><h3><strong>Dynamic Delivery’s Solution →</strong></h3><p>Dynamic Delivery provides a comprehensive platform for managing the entire lifecycle of Snowflake objects, including databases, schemas, tables, views, and more, within a robust CI/CD framework. Key features include:</p><ul><li><strong>Automated Deployment: </strong>It automates the provisioning, testing, validation, and deployment of Snowflake objects, significantly accelerating data delivery pipelines.</li><li><strong>Version Control and Governance: </strong>By integrating with Git-like environments, it enables version control of Snowflake object definitions and enforces policies and workflows for controlled and auditable deployments.</li><li><strong>Reliable Design and Deployment: </strong>Dynamic Delivery facilitates reliable design and deployment through templating, sandboxing capabilities, and AI-assisted guidance for optimal object design and functionality.</li><li><strong>Automated Testing: </strong>It incorporates automated testing to ensure the quality and integrity of Snowflake objects throughout the CI/CD pipeline.</li><li><strong>Simplified Setup: </strong>As a Snowflake Native App, Dynamic Delivery integrates seamlessly with Snowflake, eliminating the need for complex third-party tools or extensive custom configurations.</li><li><strong>AI Copilot Integration: </strong>The solution includes an AI Copilot for features like SQL generation from natural language, DDL script creation, and intelligent suggestions for safe object editing, further enhancing developer productivity.</li></ul><h3><strong>Benefits for Snowflake Users →</strong></h3><p>By leveraging Dynamic Delivery, Snowflake users can achieve:</p><ul><li><strong>Faster Data Delivery:</strong> Accelerate the deployment of data products and infrastructure changes.</li><li><strong>Reduced Errors:</strong> Minimize manual errors through automation and robust testing.</li><li><strong>Improved Governance:</strong> Gain greater control and auditability over Snowflake object deployments.</li><li><strong>Enhanced Developer Productivity:</strong> Streamline development workflows and empower data engineers with advanced automation and AI assistance.</li><li><strong>Seamless Integration:</strong> Leverage a native Snowflake experience for CI/CD, simplifying management and reducing overhead.</li></ul><h3><strong>Conclusion →</strong></h3><p>In essence, Dynamic Delivery from <a href="https://www.dataops.live/">DataOps.live</a> empowers organizations to transform their Snowflake object management into a modern, agile, and efficient DataOps practice, critical for delivering trusted and valuable data products at the speed demanded by today’s data-driven enterprises.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8be576f197cb" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>