ChatGPT has become one of the fastest-growing products of all-time, and chances are people in your organization are already using it.

But are they using it the right way?

Are they setting you up for success? Or exposing you to unnecessary risks?

In this guide we're going to walk you through the pitfalls you need to be aware of to avoid liability and embarrassment. We'll also talk about some of the best use cases to help your organization perform more effectively when it comes to data analysis.

First, if you think people in your organization aren't using chat-based large language model (LLM) tools like ChatGPT, think again.

In late 2022, ChatGPT became the fastest ever product to reach 1 million users, in just 5 days:

Source: company announcements via Business Insider/LinkedIn.
Image credit: statista

Since then, Google released Bard, their chat-based LLM tool, and the arms race is on.

These tools are free, easy to use, and they can be powerful.

So it's relatively safe to assume your competitors are using these AI tools, and your company's own employees likely are as well.

If your organization is already using these tools, the first thing you want to think about is the risks being introduced and how you can mitigate them.

There are two major landmines to avoid, and a few other more minor pitfalls you should keep in mind as well. Let's get into them:

Problems with LLM tools like ChatGPT and Google Bard

1) Your data isn't private.

This is probably the biggest issue. The tradeoff you have to consider when using free tools like ChatGPT and Google Bard is that any data you upload is leaving your hands.

These tools are not private, and you should assume that anything you upload into them is being saved and could be accessed by the creators of the tools at any time.

Are they manually digging through everyone's chats? Of course not. But it's not worth taking the risk that some of your sensitive information gets out because of improper use of these tools.

You need to make sure that your organization knows not to upload any kind of sensitive information that might get you in trouble.

You don't want anyone uploading personally identifiable customer information, giving away any information about your database schema, or sharing any other proprietary strategic knowledge.

This might sound obvious, but it's still worth having the conversation with your team.

2) LLM tools are known to "hallucinate".

This is a nice way of saying they make things up... tell lies... fabricate information.

In my opinion, this is by far the largest current limitation of these chat-based AI tools. Using them requires a good deal of fact checking and critical thinking skills. You can't trust what they tell you, at least not today.

Again, even if this seems obvious to you, there might be someone on your team who doesn't understand this, and takes everything the LLM tools produce as gospel.

It's worth having a conversation with them.

If you're not careful, and rely on false information, your organization could end up embarrassed. Or worse, you could get into serious trouble, like the attorneys who used ChatGPT to prep their case, and cited fictitious cases in a court of law. You can see that full story below:

Yikes... don't be like these folks, and don't let someone on your team become them either. Be smart about how you use ChatGPT and other LLM tools, and understand the risks and limitations.

These first two are the biggest issues. If you take away nothing else from reading this, please remember these two pitfalls.

Here are a few more minor issues to keep in mind RE: ChatGPT...

3) You might get an answer to your question, but it might not be the best answer.

You might get an answer that sounds legitimate, or maybe you generate some code that seems to run. But, it might not be the best way to solve the problem.

So again, be skeptical, use the tools, but bring your own sense of judgment to the table and think critically. And make sure the folks on your team are thinking this way too.

4) Today's most prominent LLM tools are extremely broad.

This makes them universally interesting and potentially valuable to anyone on the planet. But the downside is they aren't specifically trained and tuned for most of the specific use cases that you might need help with.

This will change over time, and we can expect to see more niche LLM tools emerge to tackle certain problems and do it better than these broader models that attempt to boil the ocean.

We're already seeing this today, and it's a safe bet to think that tons of these niche players will emerge over the next 12 months.

The key takeaway here is to keep your head up and watch for tools that might be custom-built for your use case in the future.

5) LLMs don't have common sense and lack basic human judgment

At least today, these models are trained using a vast dataset of information with varying degrees of accuracy and less quality control than would be ideal.

Right now, the tools will answer your question, but they make no attempt to assess whether or not it's an objectively good answer. That's on you.

Again, make sure your people know this, and are aware they need to bring their own critical thinking skills to the table if they want to effectively use these AI tools.

How to handle the problems with LLM tools

In general, make sure you and your team know to be skeptical, bring your own judgment, and quickly learn which of your use cases tend to be good fits for AI and for which ones the tools aren't quite there yet.

If you can keep the limitations in mind, be careful with sensitive information, and use AI tools where they are most effective, they can really help speed up certain aspects of your workflow.

Also, keep in mind that these things are evolving rapidly. If you are using ChatGPT or another LLM chat tool a year or two from now, you may find some of the biggest problems have been solved. Consider this a moving target.

Best practices for using AI tools

Throughout the rest of this writeup, we'll discuss best practices and specific use cases for our favorite domain area, data analytics.

Note though that a lot of the tips, especially around prompt engineering, can be applied universally, not just to analytics.

First, let's talk about Prompt Engineering; the practice of creating prompts that will generate effective and accurate responses.

You can see some of our best practices noted below, which we'll elaborate on.

1. Be clear and specific.
With more detailed information, the AI tools are better able to give you a response that will solve the problem you're tackling.

2. Provide context.
By sharing information about your situation and the problem you are trying to solve, the AI will better understand your perspective and will tailor the response to your needs.

3. Establish roles.
We like to tell the AI who they should impersonate, and who will be consuming their response. This helps them shape a response with appropriate detail.

4. Set the tone.
You can prescribe things like how formal, how technical, and how long you want the response.

If you keep these four best practices in mind when writing your prompts, it will really go a long way.

The only final thing we should add is that you shouldn't worry about getting it perfect. Aim for pretty good, review the result you are seeing, and then iterate if needed.

Prompt engineering should be a task that a couple of minutes. Don't overthink it.

Walk through these best practices with your team, and I bet they will thank you for it.

Best Analytics use cases for ChatGPT

There are lots of great ways to use ChatGPT and other LLM tools to improve the analytics workflow. Here are some of our favorites:

• Troubleshooting or debugging your code

• Adding human readable comments to your code

• Generating code, queries, or formulas, from scratch!

• Performance optimizing code, queries, or formulas

• Providing data visualization tips

• Generating data samples

• Automating manual tasks

• Generating step-by-step tutorials

• Explaining a technical concept

Like I said, there are lots more valuable use cases. These are the ones we're seeing come up the most.

Next, we can get into some more concrete examples of how you can use ChatGPT for data analytics with tools like Excel, Sheets, SQL, Python, and Power BI.

Using ChatGPT with Excel

In this example, you can see how ChatGPT can help you understand what's going on with a particular Excel function:

Here are some other great ChatGPT use cases for Excel:

• Explaining how an Excel formula works(pictured above)

• Generating formulas from scratch (video below)

• Creating DAX or M code

• Generating VBA scripts or automation

• Troubleshooting errors in formulas or code

• Data prep & exploratory data analysis

• Generating sample data

The video below gives a detailed walkthrough, where Chris shows us how you can create an Excel formula from scratch using ChatGPT:

Using ChatGPT with Google Sheets

In this example, similar to the Excel video example, Enrique shows us how to create Google Sheets formulas from scratch

Some of our other best ChatGPT use cases for Google Sheets include:

• Generating formulas from scratch (pictured above)

• Adapting Excel tools for Google Sheets (see video below)

• Explaining formulas

• Troubleshooting errors

• Applying formatting

• Writing Regex patterns

• Coding apps scripts

The video below is pretty useful and one of our favorites. Enrique uses ChatGPT to adapt tools built in Excel to work in Google Sheets.

Using ChatGPT with SQL

Here's an example of using ChatGPT to debug errors in your SQL code:

In general, I've found SQL use cases to be useful at times, but far from perfect. The solutions will get better, but for now, you really do need basic SQL skills to make tools like ChatGPT useful. With that caveat in mind, here are some decent use cases:

• Debugging errors in your query (pictured above)

• Comment SQL code (video below)

• Explaining a SQL concept

• Describing what a SQL query is doing

• Generating SQL queries from scratch

• Performance optimizing your query

Here's a video where I walk through using ChatGPT to add human readable comments (your coworkers will thank you for good comments):

Using ChatGPT with Python

Here's an example generating Python code from scratch with ChatGPT:

Some of our favorite Python use cases include:

• Generating Python code from scratch (pictured above)

• Creating data visualization code (video below)

• Explaining Python code

• Troubleshooting errors in your code

• Performance-optimizing your code

• Researching libraries

• Web scraping

• Interpreting Machine Learning models

• Jupyter Notebooks plugin (requires a paid version of ChatGPT)

Here's a video where Chris takes us through creating some data visualization code for Python using ChatGPT:

Using ChatGPT with Power BI

Here's an example of using ChatGPT to understand how to make data connections using Power BI:

Here's a list of some of our favorite Power BI use cases:

• Connecting to data sources (shown above)

• Creating DAX calculations (video below)

• Understanding how to get started

• Explaining specific Power BI concepts

• Explaining what code is doing

• Generating measures

• Troubleshooting errors

• Creating visuals and dashboards

In the video below, Aaron walks us through using AI tools to create custom DAX measures for Power BI:

Learn More: Free Course - ChatGPT for Data Analytics

If you enjoyed reading this one, and you think you or someone on your team might want to learn more, Maven Analytics just launched a brand new course: ChatGPT for Data Analytics.

In the course, we introduce you to the world of deep learning and generative AI, explore the rapid rise of large language models like ChatGPT and Google Bard, and get you up and running with free tools that will take your skills to the next level.

These are just some of the topics we’ll be covering…

• Why AI for Data Analytics?

• Intro to AI, LLM’s, & ChatGPT

• Prompt Engineering

• ChatGPT for Excel

• ChatGPT for Google Sheets

• ChatGPT for Power BI

• ChatGPT for SQL

• ChatGPT for Python

• …and more!

The course is beginner-friendly, and designed for anyone who wants to leverage modern technology to work more efficiently, and make smarter, data-driven decisions.

We are offering this course for FREE. You do not need a paid account to access the material. Just get in there and start learning!

Course Details: ChatGPT for Data Analysis

Meet Maven for Business - Train Your Team

Maven For Business is the fastest, most effective way to empower your team with expert-level data analytics skills.

For leaders, that means working with expert learning guides and modern, flexible tools to build the perfect plan for your team. Assess and index your teams’ skills, discover self-paced courses to close the gaps, and create personalized paths to help employees learn at their own pace and develop the skills they need most.

For employees, it means having a clear path to mastering tools like Excel, SQL, Power BI, Tableau and Python, and the skills to work smarter and deliver real business impact.

Upskill faster with project-based courses and case studies, learn from top instructors and industry experts, and deliver results you can measure with real-time reporting and best-in-class progress tracking.

Named by USA Today as one of the top 10 education companies revolutionizing the industry, Maven Analytics has helped more than 1,000,000 people around the world build world-class data literacy and analytics skills.

Learn more about Maven for Business below:

LEARN MORE: Maven For Business