How to Do Agentic Data Analysis with Datacamp
Dave Wentzel
linkedin.com/in/dwentzel
Who is Dave?
I work for Microsoft's Innovation Hub in Malvern PA as a data scientist. The MIH helps customers solve business problems using our technology platform. Dave’s concentration is in analytic solutions and shortening time-to-value by leverage Azure and a modern approach to analytics using Lean methods and Design Thinking.
See also: nl2sql: "Natural Language to SQL....the first phase in conversational analytics."
Vibe Analytics empowers business professionals to have intelligent conversations with their data—without writing code. But it's not just for business professionals. IT folks can leverage these techniques to accelerate time-to-market for analytics solutions.
Every day, business leaders need to make critical decisions based on data. But accessing those insights typically requires:
- Submitting requests to overburdened IT or data science teams
- Waiting days or weeks for static reports and dashboards
- Learning complex SQL, Python, or business intelligence tools
- Hiring expensive consultants or data analysts
By the time you get your answer, the business question has often changed.
Vibe Analytics transforms how you interact with data by using AI to guide you through a natural, conversational analytics process—similar to how you'd work with an expert data analyst, but instant and on-demand. Don't think that the AI/LLM will give you the answers, instead, view the AI as a tool you can use to reason through the business problem.
What makes it different:
- Have conversations with your data using plain language—describe what you want to know, not how to query it
- Get guided insights as AI walks you through the entire discovery process, from data quality checks to actionable recommendations
- Move beyond static reports to dynamic exploration that adapts to what you're learning about your data.
- Make confident decisions backed by AI-assisted analysis that helps you avoid common analytical pitfalls and cognitive biases
Instead of just answering "what happened," Vibe Analytics helps you discover:
- Hidden patterns in your data that traditional dashboards miss
- Prescriptive actions—not just "here's what happened," but "here's what you should do next"
- Root causes behind the trends you're seeing
- Predictive insights about what's likely to happen next
Think of it as having an expert data analyst working alongside you, available 24/7, helping you ask better questions and discover insights you didn't know to look for.
The secret behind Vibe Analytics is following the same proven process that expert data scientists use—called Exploratory Data Analytics (EDA)—but with AI as your guide so you don't need years of training.
1. Understand Your Data
- Automatically identify patterns, anomalies, and relationships
- Detect data quality issues before they lead to wrong conclusions
- Visualize trends in ways that make sense for your business questions
2. Avoid Common Pitfalls
- Stay objective by checking your assumptions against the actual data
- Identify outliers and anomalies that might skew your analysis
- Recognize when you're falling into cognitive biases (like confirmation bias or recency bias)
3. Clean and Prepare Data
- Handle missing or inconsistent data intelligently
- Merge data from multiple sources
- Transform messy data into analysis-ready insights
Why This Matters: Even the best business minds can reach wrong conclusions with data if they skip these critical steps. AI ensures you follow best practices automatically, without needing to become a data scientist yourself.
The demos in this repo will cover all of the above topics.
Using Vibe Analytics is as easy as having a conversation. Here's the process:
1. Start with Your Business Question
Describe what you want to understand in plain language using the CRIT framework:
- Context: What's the business situation?
- Role: What is the persona you want the LLM/AI to play? (data scientist, data engineer, marketing analyst)
- The more specific you are, the better the conversations will be. Examples:
- You are a data scientist with 10 years of experience in manufacturing...
- You are a data analyst, skilled in Exploratory Data Analysis, statistics, and Design of Experiments (I have other examples throughout the repo)
- Interview: Tell the AI the role you play and tell it to "interview me", we want to have a conversation about the problem.
- This solves the sophistry and sycophancy problems with LLMs. While we want the LLMs to overcome OUR biases, we also don't want to introduce the LLMs possibly biases into our reasoning.
- One sure-fire way to ENSURE an LLM doesn't steer you in the wrong direction is to always say something like "This is my problem, give me 5 ideas on how to solve it and we will explore them together."
- I call this thinking a meta-layer higher.
- Task: What decisions do you need to make?
2. Collaborate with AI
- Review the insights and visualizations AI generates
- Ask follow-up questions as new patterns emerge
- Refine your understanding iteratively—just like brainstorming with a colleague
3. Rinse and Repeat
Continue the above interview process until you are satisfied with the result.
This is exactly how business leaders should work with data—starting with business questions, not database schemas. The AI handles the technical complexity while you focus on business insight and decision-making. The best data scientists/consultants have been doing it this way for decades.
From Static Reports to Dynamic Discovery:
- Traditional dashboards show you what happened yesterday; Vibe Analytics helps you understand why it happened and what to do about it.
- Dashboards and traditional reports have built-in biases...they present the data to the business person based on how the developer designed them.
- Transform one-way reporting into two-way conversations where each answer leads to better questions.
- This is EXACTLY what business people have wanted for DECADES. True self-service analytics.
From Descriptive to Prescriptive:
- Move beyond "sales were down 15%" to "sales declined because of X, Y, and Z—and here are three actions you should take immediately"
- Get AI-guided recommendations based on patterns in your specific business data
- If you also give your LLMs access to your business knowledge that is locked up in pdfs and Sharepoint lists, the output is even better.
Good news: You don't need to be a programmer or data scientist. Here's what you need:
The easiest way to start with Vibe Analytics today:
-
VS Code (free code editor from Microsoft)
- Don't worry—you won't be writing code manually
-
Microsoft Fabric (cloud data platform)
- Provides a cost-effective workspace for testing and personal use
- Includes built-in data tools and visualizations
- Note: You don't need "Fabric Copilot"—regular GitHub Copilot works great
-
GitHub Copilot (AI assistant)
- Works inside VS Code to understand your questions and generate analysis
-
Jupyter Notebooks (interactive data environment)
- Think of it as a smart canvas where you ask questions and see visual answers
- Handles charts, graphs, and data connections automatically
The platform is flexible and works with multiple technologies:
- IDEs: VS Code (recommended), Jupyter environments, Claude Code...any IDE will work if it can connect to your data, leverage a Jupyter notebook, and interact with an LLM agent.
- Compute engines: Python, SQL, or Apache Spark with MCP server support
- AI models: GitHub Copilot, Azure OpenAI, Anthropic/Claude or other coding agents
- Data platforms: Microsoft Fabric, Azure, SQL Server, Oracle, Excel or other cloud/on-premises data sources. If you can connect to the data it will work.
Wait! I'm a business person and I don't understand any of this. How can I get started quickly. I know where my data resides within my company (or I'm using Excel files). See this alternative setup for business people
I usually hear something like this from IT people:
This is really cool Dave but there is NO WAY I'm giving vscode to my executives.
First, you'd be surprised how many of your execs are currently using Claude Code or vscode in their daily jobs.
However, while these demos use VS Code, enterprise deployments can provide browser-based access so business users never need to install development tools. Contact the Microsoft Innovation Hub if you're interested in learning how to set up Vibe Analytics for your entire team.
In a nutshell, we can vibe code an appropriate web-based UI/UX for your people in less than an hour.
These demos show actual business scenarios where Vibe Analytics delivers insights that would be difficult or impossible with traditional tools.
See how to use natural language questions and the CRIT framework to explore data, uncover insights, and get actionable recommendations—without writing any code.
- Complete walkthrough: CRIT prompting + Guided Analytics
- Real example using Microsoft Fabric
- Shows the complete process from business question to decision
Can AI spot subtle data problems that even experienced analysts miss? Can you? This demo reveals common analytical pitfalls and shows how AI helps you avoid costly mistakes. Hint: you may need to look at the visualizations that are created to spot the issues.
If you don't have Fabric and want to follow along. This will show you everything you need to reproduce this demo in your environment.
Interactive Demo: Cognitive Mistakes in Data Analysis
**If you don't have Fabric and want to follow along
See how to analyze customer survey data to uncover sentiment patterns, preferences, and actionable marketing strategies—all through conversation with AI.
Marketing Survey Analytics Demo
Don't have fabric? Follow this alternate setup, which also covers the gist of the demo.
Turn messy, poorly structured Excel files into powerful insights. Watch AI clean, restructure, and analyze data in ways that Excel alone can't handle.
- Takes real-world "Excel chaos" and transforms it into clean, actionable insights
For technical teams: See how Vibe principles apply to data engineering tasks like real-time streaming data processing—all with minimal manual coding.
Vibe Data Engineering with Spark Streaming
A manufacturing OEE (Overall Equipment Effectiveness) story for discrete manufacturing (blow-molded HDPE bottles). The obvious answer—"replace the bad press"—turns out to be wrong once the LLM helps us reason across three disconnected systems (SAP MII, an MES/PI historian, and a process engineer's Excel file). One root cause explains all three OEE losses, and we avoid a $500K capital mistake.
Manufacturing OEE Root-Cause Demo
- Shows how LLM-assisted analytics stitches together data no single dashboard connects
- Great for plant leadership: the deliverable is a dollar figure, not a chart
Vibe Analytics represents a fundamental shift in how business professionals interact with data—from passive consumers of reports to active explorers of insights.
- Business Analysts who need to answer complex questions without waiting for IT
- Department Leaders who want to make faster, data-driven decisions
- Product Managers seeking to understand customer behavior and market trends
- Marketing Teams analyzing campaign performance and customer insights
- Finance Professionals exploring trends and forecasting scenarios
- Anyone who has ever been frustrated by the gap between having a question and getting an answer
- Try the demos above to see what's possible
- Review the setup guide to understand the technical requirements. If you're non-technical, try the alternate setup guide
- Contact Microsoft Innovation Hub to discuss enterprise deployment options