3 High-Quality Prompts To Use ChatGPT for Better Analytics

Hi there,

In today’s newsletter, we'll dive into the world of AI for analytics with a special focus on the world’s most popular AI application - ChatGPT.

We'll discuss how it can complement your data analysis process, even without sharing confidential data.

I'll share three techniques you can use to effectively leverage ChatGPT and add value to your role as a data professional.

Let’s go!

🤿 Want to dive deeper? Check my upcoming AI LIVE Workshops to take your AI skills to the next level to innovate and grow your business. 🚀

ChatGPT is NOT the better data analyst, but…

A common misconception about ChatGPT is that it will replace human jobs.

But in fact, ChatGPT is a tool designed to work alongside you, augmenting your ability to perform your tasks efficiently.

Instead of doing the job for you, ChatGPT works best when doing the work with you – getting faster outputs, improving quality, and serving as a second pair of eyes to eliminate blind spots.

That’s why I don’t recommend using prompts like "Here's my data in CSV, tell me what's going on!". These use cases are flawed for three reasons:

1. The need to expose confidential data

2. The inability to verify the correctness without double-checking

3. You miss the chance to get familiar with your data and really understand what's going on.

Therefore, the most powerful use of ChatGPT in data analysis is to use it before data analysis begins. How?

Let me walk you through three practical examples:

1. Crafting SMART Problem Statements

The most critical task of a data analyst is to ask the right questions.

That's why a robust data analysis project should start with a well-formulated problem statement. Here's where the SMART (Specific, Measurable, Actionable, Relevant, Time-bound) principle comes in.

ChatGPT can aid in transforming a non-SMART problem statement into a SMART one by asking targeted questions that can provide you a well-rounded perspective.

For example, let's say start with a loose problem statement like this:

"How can I improve my sales performance?"

This is actually not a problem statement, but an undirected question. It's very hard to do any analysis here. But with the help of ChatGPT, we can turn this loose question into an actionable SMART problem statement that looks like something like this as a final output:

This problem statement is now Specific (focused on marketing to increase MRR), Measurable (aiming to increase from $50k to $100k), Actionable (through new marketing strategies), Relevant (aligning with the business goal of growth), and Time-bound (to be achieved in the next 6 months).

So how did we use ChatGPT to get here?

Prompt Example for SMART Problem Statements

To turn your vague problem description into a SMART problem statement, you have to tell ChatGPT four things:

• The role: Who should it act like?

• The task: What should ChatGPT do?

• The instructions: How should ChatGPT do it?

• The context: What does ChatGPT need to know to complete the task?

All in all, here's how that first prompt could look like:

As you can see, this base prompt is quite long, but that’s ok. ChatGPT will ask you different questions and provide suggestions to help you refine your problem statement until you have a SMART problem statement:

Feel free to check out this link for the full conversation and easy copy/paste of the prompts to customize it to your needs: ChatGPT prompt for SMART problem statements

Note: I used GPT-4 for this and all following examples.

2. Structuring Your Analysis With Issue Trees

Now that we have a SMART problem statement, what do we do with it?

Issue trees are a powerful tool for visually representing and breaking down complex problems into smaller, more manageable chunks. Used extensively by companies such as McKinsey and BCG, issue trees provide a great graphical breakdown of your problem, making it much easier to manage and also allowing you to easily share your thought process with others.

Here’s how a great issue tree looks like (credits to this awesome issue tree tutorial):

In short, here's how to create an issue tree:

1. Start with a (SMART) problem statement: The main problem or question that needs to be answered or solved.

2. Divide problems into sub-problems and components: Each sub-problem or component should represent a different aspect or approach to the main problem.

3. Create branches for actions, questions, or topics: These will guide the exploration of each sub-problem or component.

The “secret sauce” of Issue Trees is the so-called MECE principle, which stands for Mutually Exclusive and Collectively Exhaustive. Is guarantees that your issue tree covers the whole solution space and that individual sub-topics do not overlap.

Although ChatGPT can't create visual representations, it can guide you in creating an issue tree that adheres to the MECE (Mutually Exclusive and Collectively Exhaustive) principle.

Here’s how the end result could look like, given our problem statement above as an input:

With a bit of PowerPoint hacking we could easily turn this into a diagram like this:

So how was ChatGPT at play here?

Prompt Example for Issue Trees

To go from a SMART problem statement to a fully-fledged issue tree, we again need our four magical ChatGPT components - the role, the task, the instructions and the context.

The context in this case is even a bit more lengthy, because creating MECE issue trees is actually not that simple.

That’s why I’ve provided ChatGPT five methods it can use to create the issue tree (algebraic structures, process structures, conceptual frameworks, segmentations and opposite words). And when we ask ChatGPT to create the issue tree not in one go but level by level, this gives us better control.

Here’s how the prompt looks like in my case (truncated):

The prompt is actually too long to be displayed here. That’s why I’ve shared the whole chat conversation with you so you can easily copy/paste this and fit your needs: ChatGPT prompt for issue trees

3. Developing Research Hypotheses

Once you have your SMART problem statement and your issue tree, you'll likely need to create a research hypothesis.

ChatGPT can help you refine your hypotheses and structure your experimental designs based on the problem statement and problem tree provided.

Let’s say we want to improve our brand awareness through traditional media and we’re not sure which of two possible ads to choose.

Here’s a research hypothesis ChatGPT would suggest to us:

See how nicely it aligns with our overall problem statement and issue tree framework? That’s the backbone of an effective data analysis.

How did we get here? Let’s find out!

Prompt Example for Research Hypotheses

In this case, I was lazy with the prompt design because it worked well without having to provide much context. Essentially, I just provided the output from the previous step (SMART problem statement + issue tree), and that's enough to generate a research hypothesis for a selected branch of the tree, given some input from me.

Here's the prompt I used:

(Check this link for the the full chat history)

As a result, ChatGPT gave me this well-crafted research hypothesis:

“Implementing a new advertisement campaign focused on 'X' via traditional media outlets will result in a significantly higher number of QR codes scanned, thereby increasing customer awareness and ultimately Monthly Recurring Revenue (MRR), compared to our current 'Y' advertisement campaign.”

Bonus: If you’re unsure how to set up the data analysis experiment for this case, feel free to ask ChatGPT as well!

To follow along, check out the following link, where I've shared the entire chat history with you:

• ChatGPT prompt for designing a research hypothesis

Wrapping up

In today's article, we’ve explored how ChatGPT can support your work as a data analyst - letting you do what you do, just faster and better.

Remember, ChatGPT is your co-pilot and you're in the driver's seat - a workflow I call augmented analytics.

With this powerful technology at your fingertips, you don't have to worry about AI replacing your job. Instead, you're the one on the edge.

As always, thanks for reading, and feel free to leave a comment or share this article with a friend if you found it useful.

See you next Friday,

Tobias

Want to learn more? Here are 3 ways I could help:

1. Check my books: AI-Powered Business Intelligence and Augmented Analytics.

2. Join my workshops: Regular LIVE sessions featuring a mix of theory and practical, hands-on exercises and use cases.

3. Follow me: I regularly share free content on LinkedIn and X.