Governed AI Analytics
Ask your data in plain English. Understand the definition behind the answer.
On this page
Give AI a foundation for better answers
“What was our margin last month—and which definition did that answer use?” Cloud Data Consulting is a Denver-based data consultancy that builds governed platforms to answer questions like this. We implement AI analytics in your Snowflake account, grounded in the business definitions your team agrees on.
In a reference build, two definitions of transaction count returned 197 and 204. Both numbers looked plausible; only one matched the agreed KPI. We address that upstream, by working with metric owners to define the calculation and making the definition visible in the answer.
Illustrative implementation
From a Question to a Governed Answer
A semantic model records the approved definitions and relationships the assistant uses. MCP is the connection that lets an approved assistant call the tools in your environment.
Define the metric
Review the calculation with its business owner and store a verified query alongside the semantic model.
Deploy with controlled access
Deploy the semantic view and agent, configure the agreed MCP access path, and validate the user’s permissions. Each engagement has its own deployment configuration.
Make the definition visible
Configure answers to identify the definition and verified query used, and flag unverified results. Test that behavior before release and after changes.
verified_queries: - name: net_sales question: Net sales last month? sql: | SELECT SUM(net_sales_amount) FROM __daily_sales WHERE sales_month = DATE_TRUNC( 'MONTH', DATEADD(month, -1, CURRENT_DATE()))
Semantic view ............. validatedAgent configuration ....... appliedMCP tool access ........... configuredUser role ................. read-onlySecondary roles ........... disabledAccess tests .............. reviewedQuestion-set evaluation ... reviewed
Question: Net sales last month?Net sales: $1,240,000Period: previous calendar monthDefinition: sales less returns and discountsVerified query: net_salesNo verified match? Label as unverified.
What We Deliver
Agents over your business model
Cortex agents over a semantic view of your modeled data. We assess existing BI definitions and agree where they belong so AI reuses agreed definitions instead of creating a competing source of truth.
Access from approved assistants
A Snowflake-managed MCP server configured for your chosen Claude or ChatGPT access path. We review authentication, connector requirements, and which results leave the account.
A per-user security model
Single-purpose identities with a least-privilege read-only role, secondary roles disabled, and privileged roles blocked at the OAuth integration. Compute controls follow the agreed architecture.
Verified KPIs and labeled answers
Core metrics encoded as verified queries, reviewed with their business owners. We configure answers to cite their definitions and flag unverified results, then evaluate that behavior.
Evaluation before release
A repeatable question set and an answer-quality baseline. The release process reruns evaluation after model or KPI changes, including ambiguous questions and questions outside the agreed scope.
Infrastructure your team keeps
Agents, semantic views, and access controls in your Snowflake account, with versioned configuration, runbooks, and training. Your team can maintain the implementation and its definitions.
The foundation is implemented for your account and agreed KPI set. Need architecture advice before choosing a package? Explore our AI Architecture & Implementation service.
Start Focused. Expand With Evidence.
Choose the questions
Start with one business area, its owners, and the recurring questions that consume analyst time.
Build the foundation
Agree definitions, configure access, and connect the chosen AI experience to the modeled data.
Evaluate before expanding
Test calculations, ambiguous questions, access boundaries, and questions the system should decline.
Learn from actual use
Review unanswered questions and failures, add verified examples, and rerun the evaluation before releasing changes.
See the method behind this approach: Building the Verified Query Flywheel.
Is Your Data Ready?
This offering fits teams with a Snowflake warehouse, modeled business data, and recurring analytical questions. You also need a business owner to approve definitions and a technical owner for access and operations.
If the warehouse still needs modeling or its metrics disagree, we scope that foundation first. Our dbt consulting and data governance services support that work.
Why Cloud Data Consulting?
We work across the warehouse, semantic model, AI integration, and evaluation process. That lets us address a wrong answer at its source: a conflicting metric, an incorrect join, a missing access rule, or a question outside the model’s scope.
The foundation uses reusable deployment patterns and a verified-query-first method, tailored to your business. Ongoing KPI expansion and support can be scoped separately. For broader program leadership, explore our Fractional Head of Data & AI offering.
Choose the Scope You Need
Foundation
Architecture, agents, MCP access, and verified-query governance for an agreed set of KPIs. Includes initial evaluation, documentation, and team training.
Managed
The foundation plus ongoing semantic-model tuning, KPI expansion, evaluation, and support as real questions surface.
Frequently Asked Questions
Does our data leave our environment?
The semantic view, agents, and access controls run in your Snowflake account. When you connect an external assistant such as Claude or ChatGPT, query results can be returned to that service. We review that data flow and your approved tools before enabling access.
How is cost controlled?
We configure warehouse sizing and resource monitors for warehouse compute, and review AI service usage separately. Warehouse resource monitors do not cap all AI charges. Engagement scope and ongoing operating costs are discussed before implementation.
How long does it take to stand up?
The scope depends on how well your warehouse is modeled, how clearly KPIs are defined, and which access paths need approval. We assess those dependencies before agreeing the delivery plan.
What happens if we stop working with Cloud Data?
The implementation and its configuration remain in your account. Your team receives the runbooks and training to operate it; platform subscriptions, maintenance, and access administration remain your responsibility.
Which assistants can we use?
We scope Snowflake-managed MCP connectivity for approved tools such as Claude and ChatGPT, subject to their account, authentication, and connector requirements. The assistant and model choices are agreed during architecture design.
We already have dashboards. Why this?
AI analytics adds a way to explore follow-up questions in plain English. We use existing BI measures as the starting point for verified queries and reconcile definitions with metric owners, so chat and reporting share the same business logic.
Which Questions Should Your Data Answer?
Bring a business use case. We’ll map the data readiness, architecture, and first implementation scope.
Schedule a discovery call