- Feb 14, 2025
- 10 minutes
Written By
-
Alex Dean
For decades, business stakeholders have pursued real-time access to data-driven insights. The vision of having information at your fingertips isn’t new, but turning it into reality has required years of technological advancements. One of the earliest breakthroughs came from Qlik with its QlikSense Server. This platform didn’t just generate reports; it integrated Natural Language Processing (NLP) into its API, setting a precedent for conversational analytics.
The Evolution of NLP in Data Platforms
In 2016, we worked on a chatbot for Service NSW that used QlikSense as the intelligence layer between users and public statistics. It could interpret table names, measures, and dimensions based on user inputs, pulling relevant data from pre-approved sources. However, ensuring responsible data use—such as preventing cross-filtering to identify individuals—was a major challenge that ultimately constrained the project.
By 2018, Microsoft introduced its QnA feature in dashboards, allowing users to query datasets in plain language. While promising, the tool often misinterpreted queries, requiring users to rephrase their prompts multiple times. This friction led to limited adoption.
Fast forward to today, and Generative AI (GenAI) has transformed how we interact with data. Here’s why it’s a game-changer:
GenAI Can Analyze Your Data Before Generating Answers
In the past, NLP-driven tools relied on basic entity extraction, often leading to misinterpretations—e.g., retrieving revenue for “2018” instead of the product line “mountain-black-2018.” Today, GenAI can analyze table schemas, semantic models, and metadata before crafting a response.
Early models like GPT-3.5 managed around 2,000 tokens, but newer versions, processing 128,000+ tokens, provide a deeper understanding of complex queries. This expanded context window allows for more precise answers.
GenAI Can Generate More Robust Code
Previously, converting user queries into SQL required pre-programmed rules engines, which lacked flexibility. Now, GenAI dynamically generates tailored SQL queries, improving accuracy and reducing errors. Leading AI models like Claude, Google’s Gemini 2, and OpenAI’s o3 are advancing SQL generation, bridging the gap between user intent and data structure.
How to Leverage GenAI in Data Platforms
Major data platform vendors are integrating AI capabilities into their ecosystems. Here’s how some of the top players are enabling AI-driven data interactions:
Snowflake – Cortex
Snowflake’s Cortex suite includes Cortex Analyst for structured data, Cortex Search for unstructured data, and the Cortex Chat API for seamless interaction. Snowflake allows customers to choose from various AI models, including its own Arctic model, Meta’s Llama 3, and Claude. With multi-cloud support (AWS, Azure), Cortex provides flexibility in data hosting.
Microsoft Fabric – AI Skills
Microsoft Fabric simplifies natural language querying with AI Agents that generate SQL queries for Fabric warehouses. However, some limitations remain:
- AI Agents only generate and execute queries—they don’t interpret data.
- No memory of previous queries, limiting contextual follow-ups.
- Only supports English (as of Dec 2024), with no Teams integration.
- Users cannot swap out the underlying AI model, as it’s locked into Microsoft’s ecosystem.
These constraints may limit deeper, iterative data exploration.
Databricks – Genie Spaces
Databricks, known for its enterprise-scale AI and ML capabilities, has introduced Genie Spaces—an AI agent enabling iterative, context-aware data interactions. Unlike other AI tools, Genie retains memory, allowing users to refine queries and engage in deeper data conversations.
Genie also enhances data visualization by rendering results in tables or charts and includes built-in evaluation tools for quality testing. However, it’s tied to Azure OpenAI, restricting model selection. On the plus side, Databricks avoids Microsoft’s abuse monitoring, ensuring interaction data remains private.
Palantir – AIP
Palantir Foundry has long used an ontology-based approach to data management. With the introduction of the Artificial Intelligence Platform (AIP), Palantir has embedded AI into enterprise workflows, enabling:
- Automated detection of critical data points and anomalies.
- Real-time data distribution via AI agents.
- AI integration into operational processes.
Unlike standalone chat-based solutions, AIP focuses on embedding insights directly into business workflows, making it a powerful choice for enterprise-scale AI adoption.
Conclusion
Generative AI has improved how businesses interact with data, addressing past limitations in NLP-driven analytics. With platforms like Snowflake, Microsoft Fabric, Databricks, and Palantir leading the way, organizations now have better tools to unlock AI-powered insights.
