Expert insight
Snowflake Intelligence: your data knows everything, it's time to listen to it and take action
An analysis of Snowflake's new intelligence layer and what it changes for your data teams.
Snowflake Intelligence: Talking to all your data, not just your data warehouse.
In November 2025, Snowflake announced the general availability of Snowflake Intelligence, presented as an “enterprise intelligence agent” that allows every employee to ask complex questions in natural language and get actionable answers, directly in Snowflake (*).
In just a few months of preview, over 1,000 customers have already deployed more than 15,000 agents in production, giving a good indication of the traction of these use cases (*).
In just a few months:
But behind "chatting with your data" lies a much deeper transformation:
Unification of structured and unstructured data within a single AI framework.
Secure conversational access to the entire information assets.
Strengthening of governance and the semantic layer, which is no longer confined to dataviz tools.
It is precisely around these three areas that we position KPC and our Talk to Data offer.
It is precisely around these three areas that we position KPC and our Talk to Data offer..
Area 1 – Unify structured and unstructured data: expanding the company's information foundation
Historically, Snowflake established itself as the reference platform for structured data (tables, metrics, facts, dimensions). With the AI Data Cloud and building blocks like Cortex AI, Document AI or Cortex Search, this promise has progressively extended to unstructured content.
It relies on unified access to structured and unstructured data, within the same governed Snowflake environment. Particularly with audio analysis.
Thanks to Document AI, it becomes possible to extract information from invoices, contracts, emails, PDF reports... and reconcile them with master data (customers, contracts, products...). (*)
Agentic analytics capabilities (Agentic Document Analytics) make it possible to ask a question that involves thousands of documents, far beyond classic RAG architectures limited to a few dozen documents per query.
* Source Snowflake (4)
Concretely, what does this change?
Until now, most data & BI projects remained centered on numbers (revenue, margin, NPS, volumes, deadlines, etc.) fed by a few core systems (ERP, CRM, business applications).
With Snowflake Intelligence, the enterprise information foundation expands:
We no longer look only at "How many customer returns?", but also "Why these returns?" by analyzing the text of tickets, emails, or NPS surveys.
We no longer just track the dispute rate, but we explore recurring contractual clauses in conflict cases.
We no longer settle for just an IT backlog, we can automatically bring to light recurring themes in business requests.
Key Takeaway: In other words, Snowflake Intelligence completes the movement started by Snowflake: making the platform not just a data warehouse, but the "datum" of the company, the gathering point for all strategic information — both structured and unstructured.
Area 2 – Revolutionizing data access: from SQL query to dialogue - Second breakthrough: the conversational interface.
Snowflake Intelligence is primarily an out-of-the-box application, with a chat interface in Snowsight, allowing a business user to ask questions in natural language and letting the agent:
Interpret the question
Generate the appropriate queries
Cross-reference the right sources
Deliver an understandable response.
This interface relies on:
Cortex AISQL and Cortex Analyst, capable of transforming a business question into SQL, then explaining or enriching the results.
Leading LLMs (OpenAI, Anthropic, Meta, Mistral, DeepSeek…) operated directly within Snowflake, via Cortex AI. URL
Cortex Agents which orchestrate complex steps (decomposing a question, selecting the right sources, applying the right tools, etc.). URL
For business teams, this means:
Less dependency on data teams for every new question.
An ability to explore, test hypotheses, dig into a "why" in real time.
An interaction mode that feels like a discussion with a colleague who is highly knowledgeable about the data, rather than a report request.
Example: “Explain the decline in margin for BU X in Q3, taking into account commercial discounts, product mix, and logistics costs, and suggest three priority courses of action.”
Snowflake Intelligence does not just output a number. It is capable of:
Finding the right tables and views,
Decomposing the problem,
Reasoning through the different factors,
Producing an argued summary, optionally accompanied by visualizations.
Key Takeaway: This revolution in data access is not just a change in UX; it is a cultural shift: data truly becomes accessible "on demand," in the language of the business teams.
Area 3 – Semantic model and governance: when AI forces you to solidify your data stack
Third area, often underestimated: Snowflake Intelligence puts the semantic layer back at the center of the game.
Snowflake Intelligence relies on a rich understanding:
of metrics (definitions of revenue, margin, churn, etc.),
of business relationships (customer, product, contract, region, channel…),
of existing BI contexts (Tableau, other dataviz tools, "manual" SQL, etc.).
Snowflake even introduces capabilities for a “semantic layer on autopilot”: the platform learns from executed queries, frequent questions, and definitions already present in your tools, and proposes to enrich and adjust the semantic model over time.
What this changes is that:
The semantic model no longer lives solely in the dataviz tool (in the form of local calculations, extracts, scattered definitions).
It moves closer to the data foundation — inside Snowflake — and becomes a governance block in its own right, shared between IT, data, and business teams.
Every response produced by Snowflake Intelligence must be traceable: we can trace back to the tables, views, calculations, and business rules, which is key for compliance and trust.
In other words, conversational AI requires making the entire data stack robust and interoperable:
Data quality and freshness,
Rights and roles management,
Clear definition of reference metrics,
Documentation and business alignment.
Key Takeaway: At KPC, we see Snowflake Intelligence as a test of the data foundation's solidity: if the answers are not reliable, it is not (only) an AI problem, it is a revealer of structural weaknesses in governance and the semantic model.
What this implies for organizations: it is not "plug & play" on the foundation side
On the user side, Snowflake Intelligence gives an impression of simplicity: you open an interface, ask a question, get an answer. On the organization side, however, the step to climb is real:
Clarify priority use cases
On which scopes do we start?
Which decisions do we really need to accelerate?
Which risks (regulatory, business, reputational) do we need to manage?
Consolidate the data assets
Ensure quality and completeness of data in the targeted domains.
Reduce integration debt, silos, and redundancies.
Structure the semantic layer
Identify "golden" metrics and their official definitions.
Eliminate discrepancies between local versions of the truth.
Document calculation rules, aggregations, and filters.
Govern the use of AI
Define user profiles and levels of autonomy.
Set up evaluation and monitoring mechanisms for responses.
Build training and change management programs.
Snowflake already provides native building blocks to evaluate agents (like Agent GPA, which automatically measures response quality on standard datasets, with announced error detection close to human performance). (*)
But for an organization, the real question is: how to integrate these capabilities into an operational, controlled framework, aligned with my business challenges?
This is where KPC comes in with Talk to Data. Discover our solution to implement Snowflake Intelligence.
What now?
Snowflake Intelligence marks an important step: the one where we can finally talk to all our data, in a governed environment, without multiplying tools or duplicating information.
But this promise is only fully realized if:
the data foundation is solid,
the semantic layer is structured,
use cases are prioritized and governed,
users are supported.
As a Snowflake partner and participant in the launch program of Snowflake Intelligence in General Availability, KPC is positioned to help organizations cross this threshold, combining:
the power of the AI Data Cloud,
the maturity of Snowflake's agentic capabilities,
and Talk to Data's expertise to turn it into a genuine driver of transformation.
If you would like to explore how Snowflake Intelligence can integrate into your data stack, your business use cases, and your governance, we would be delighted to discuss it.
" That's it, data is finally free. Asking a question in natural language, getting an actionable answer, deciding faster: it is already a reality. "
Article written by Mickael Kuentz, - Data & AI Director at KPC.
