Choosing an analytics platform for its AI features means weighing where the model runs, what it can reason over, and how the output is governed, not which vendor demos best. The bar moved sharply in late 2025: Snowflake's Cortex AISQL functions, including AI_COMPLETE, AI_EXTRACT, AI_TRANSLATE, and AI_SENTIMENT, reached general availability in November 2025 and run over text, image, audio, and video directly in SQL with no external API call. Databricks shipped Agent Bricks and Mosaic AI for custom agents and tuned models. This ranking scores eight platforms on native AI functions, natural-language querying, vector and unstructured-data support, and governance of AI output rather than dashboard count or visual polish.
Mosaic AI, Agent Bricks for multi-step agents, managed vector search, fine-tuning, and deep MLflow LLMOps make it the strongest base for teams building custom AI on their own data. The trade-off is that realising this depth needs data-engineering maturity. Read full review →
Cortex Analyst, Copilot, Cortex Search, and AISQL functions that reached GA in November 2025 bring AI to SQL users without external calls. The most approachable native AI for analytics teams already on the platform, with consumption cost to monitor as usage scales. Read full review →
Copilot in Fabric generates measures, narratives, and report content over governed semantic models, and pairs naturally with Azure data services. Best for Microsoft-aligned organisations, though the strongest AI features sit behind Fabric capacity rather than the base licence. Read full review →
4.5Editorial score
From $14/user/moMicrosoft estates
4
Google BigQuery (Gemini)
BigQuery ML plus Gemini-assisted SQL and data-agent features keep AI close to a serverless warehouse. A strong fit for Google Cloud data estates, with capability tied to staying inside the Google ecosystem. Review pending
Search-led analytics with the Spotter AI agent for conversational questions over governed data, suited to business users who want answers without writing SQL. Narrower than a full data platform and usually deployed alongside a warehouse. Read full review →
Qlik Answers and AutoML add generative search and predictive modelling on top of the associative engine, fitting teams that value exploratory analysis. Its AI roadmap trails the warehouse-native leaders on depth of large-model integration. Read full review →
Conversational analytics over the LookML semantic layer gives consistent, governed answers, a good fit where a single source of metric truth matters. Modelling discipline is required before the AI layer pays off. Read full review →
An AI service layer and low-code app building suit organisations that want packaged data apps with embedded assistance. Less of a foundation for custom model development than Databricks or Snowflake. Read full review →
4.2Editorial score
Consumption (quote)App-led BI
Selection criteria for AI features in analytics
The first criterion is where the model runs relative to the data. Warehouse-native AI, such as Snowflake Cortex AISQL or BigQuery ML, keeps governed data in place and avoids external API exposure, which matters for regulated data. Platforms that call out to external models can be more flexible but add a data-movement and governance question.
The second criterion is the audience. Analyst-facing copilots (Snowflake Copilot, Power BI Copilot, ThoughtSpot Spotter) lower the barrier for business users, while builder-facing toolkits (Databricks Mosaic AI and Agent Bricks) suit teams creating custom agents and tuned models. The third is semantic grounding: AI answers are only trustworthy when they sit on a governed model or metric layer, which is why LookML, Power BI semantic models, and certified Snowflake views matter as much as the model itself. The fourth is cost shape: most AI features run on consumption, so usage can scale faster than expected and needs monitoring from day one. A real limitation across the category: generative output can be confidently wrong, so human review and grounding remain mandatory. See the data analytics category, our best data analytics for enterprise ranking, the Databricks vs Snowflake comparison, and the best BI for enterprise guide.
Comparison on what matters for AI features
Platform
Signature AI feature
Best audience
Rating
Databricks
Mosaic AI, Agent Bricks
AI builders
4.6
Snowflake
Cortex AISQL (GA Nov 2025)
SQL analysts
4.5
Power BI / Fabric
Copilot over semantic models
Microsoft estates
4.5
BigQuery
Gemini-assisted SQL, BQML
Google Cloud estates
4.5
ThoughtSpot
Spotter search agent
Business users
4.4
Looker
Conversational on LookML
Governed metrics
4.2
Recommendation
For teams building custom AI agents and tuned models on their own data, Databricks is the strongest base, given data-engineering maturity. For analytics teams that want native AI without leaving SQL, Snowflake Cortex is the most approachable, with its AISQL functions now generally available. Microsoft-aligned organisations should lead with Power BI and Fabric Copilot, and Google Cloud estates with BigQuery and Gemini. Where the goal is letting business users ask questions in natural language, ThoughtSpot or a copilot on a governed semantic model fits best. Whichever platform you choose, ground AI output on a governed model and keep human review in the loop, because consumption cost and confident errors are the two recurring risks.
Which analytics platform has the strongest native AI?
Databricks and Snowflake lead for native AI. Databricks Mosaic AI and Agent Bricks suit teams building custom agents and tuned models, while Snowflake Cortex brings AI functions to SQL users, with its AISQL set reaching general availability in November 2025. The better choice depends on whether your team builds models or mainly queries data.
Do AI analytics features keep data in place?
Warehouse-native features such as Snowflake Cortex AISQL and BigQuery ML run inside the platform without external API calls, keeping governed data in place. Platforms that route to external models add flexibility but introduce a data-movement and governance question, which matters most for regulated or sensitive data.
Can business users use these AI features without SQL?
Yes. Copilots and search agents such as Snowflake Copilot, Power BI Copilot, and ThoughtSpot Spotter let business users ask questions in natural language. Their reliability depends on a governed semantic or metric layer underneath, so investment in modelling is what makes natural-language answers trustworthy.
How much do AI analytics features cost?
Most AI features run on consumption pricing on top of the base platform, and some sit behind a higher tier or dedicated capacity, such as Microsoft Fabric. Costs can scale faster than expected as adoption grows, so monitor usage from the start and set guardrails rather than assuming a flat licence cost.
What is the main risk of AI features in analytics?
The main risk is confident but incorrect output. Generative answers can look authoritative while being wrong, so grounding on governed data and keeping human review in the workflow are essential. Treat AI features as an accelerator for analysts rather than an unsupervised replacement for validated reporting.
Last updated:
Get a free, independent vendor shortlist
Tell us what you're evaluating and we'll send a tailored shortlist of vendors that actually fit — no vendor funding, no pay-to-play.