Top 9 High-ROI ChatGPT Use Cases for Product Managers
How to master ChatGPT as a Product Manager and save 8+ hours a week.
Hey, Paweł here. Welcome to the premium edition of The Product Compass.
Every week, I share actionable insights and resources for PMs.
Consider upgrading your account, if you haven’t already, for the full experience:
ChatGPT has 180+ million users.
It's crazy what results PMs can get with ChatGPT-4o. But just a few write good prompts.
If you are a senior Product Manager, an LLM won’t generate better answers. But, when prompted correctly, it can do 80% of the job. And let you focus on what matters most.
So, in today’s newsletter:
How to Choose and Access the Right LLM
Top 9 AI Prompting Hacks
Top 9 High-ROI LLM Use Cases for Product Managers:
3.1 Came up with new ideas
🔒 3.2 Identify hidden assumptions
🔒 3.3 Plan the right experiments
🔒 3.4 Summarize a customer interview
🔒 3.5 Summarize a meeting
🔒 3.6 Boost social listening (sentiment analysis)
🔒 3.7 Write good user stories
🔒 3.8 Generate SQL queries for data analysis
🔒 3.9 PRD (product requirements document) and other templates
1. How to Choose and Access the Right LLM
Each available LLM (Large Language Model)* performs differently at different tasks. And you can compare them using dozens of metrics.
But I have learned that those metrics do not necessarily translate into user experience.
So, how do you choose the right LLM?
The most reliable source of information I found is the LMSYS Chatbot Arena Leaderboard. The scores are assigned by humans (currently 800,000+) who compare random models in A/B tests.
The current leaderboard (5/25/2024):
I’m often puzzled that many PMs continue to use ChatGPT-3.5 Turbo (1121 points), which doesn’t even make it into the top 25. There are many better, cheaper, and faster alternatives.
Models I encourage you to consider at work:
ChatGPT-4o (1287 points)
This is the best available model, released on May 10, 2024. Interestingly, it will get interactive tables and charts and new multimodal features (the latter are currently available through the API) in a few weeks.
Price: $20/month
Get access: https://chat.openai.com/
ChatGPT-4 Turbo (1262 points)
This is the second-best model you can access for free since March 2024 via Microsoft Copilot in Bing. I expect Microsoft to use Chat GPT-4o soon, as it’s simply 50% more cost-effective.
Price: Free
Get access: https://www.bing.com/chat
Google Gemini 1.5 (1248 points)
Available only in the Gemini Advanced plan for $20/month. In the free version, you can use Gemini Pro (1208).
Price: $20/month
Get access: https://gemini.google.com/app
Claude 3 Opus (1246 points)
While Claude is good for philosophical discussions, it’s less useful when structuring data. It’s also not available in many countries. Workarounds require a US phone number and a VPN.
Price: $20/month
Get access: https://claude.ai/
Llama3 70b (1203 points)
An open-source model from Meta. You can access it for free on the Groq website, together with a few weaker open-source models. This model is extremely fast.
Price: Free
Get access here: https://groq.com/
*I use LLM (Large Language Model), as this is what people commonly call AI models, but, in fact, many of them are LMM (Large Multimodal Models). Please note that the LMSYS Chatbot Arena Leaderboard rates only the text responses to text questions.
2. Top 9 AI Prompting Hacks
Choosing the right model and asking the right questions is not enough; you need to ask those questions right.
Here are nine techniques to help you get the best responses from your LLM:
2.1 Ask to play a role
Assigning a role AI has to play can make the response more relevant and aligned with your use case:
Before: Summarize this <document>.
After: Imagine you are a senior product marketer with 10+ years of experience. How would you summarize this <document>?
2.2 Clarify the context
Providing context within your prompts helps the AI understand the situation better.
Before: Generate ten ideas on how to solve <problem>.
After: We are performing continuous product discovery. One of the identified opportunities in our <product> is <problem> for <customer segment>. Generate ten ideas that align with our <objective>.
2.3 Write like you’re talking to a human
Sometimes, it helps to write your prompts as if you’re having a conversation with a person. This makes the prompt clearer and allows the AI to respond more naturally.
Before: Generate summary.
After: Summarize the main issues discussed in this report.
2.4 Set clear expectations
Be explicit about what you need from the AI. What does a good answer look like? Are there any constraints or limitations?
Before: What are the latest trends in the <industry>?
After: Provide a detailed analysis of the latest trends in the <industry>, focusing on new technologies and market shifts in the US over the past year. Limit your response to three key trends that might have the highest impact on <product>.
2.5 Provide an example or template
Including examples or templates in your prompts can guide the AI to produce responses that better match your expectations.
Before: Create a set of user stories for <a new feature>.
After: Here are two examples of user stories I wrote in the past: <examples>. Create a set of similar user stories for <a new feature>.
2.6 Provide step-by-step instructions
This technique is also known as the “chain of thought.” You guide the model to think step-by-step through complex problems or questions.
Before: Conduct a competitive analysis for the <product> based on <market data>.
After: Imagine you are a senior product manager responsible for the <product>. First, identify the top five competitors based on <market data>. Then, analyze their value proposition based on <value proposition template>. Finally, summarize how our product compares using a <value curve template>.
Note: I combined several techniques. Can you identify them?
2.7 Avoid leading questions
Just like when interviewing customers, it’s essential not to suggest the expected answer. Doing so might easily bias the AI’s response.
Before: Why is <product> better than its competitors based on <data>?
After: What are the strengths and weaknesses of <product> compared to its competitors based on <data>?
2.8 Raise the stakes
Framing the task as high-stakes or important can encourage the AI to “try harder.”
Before: Provide a detailed analysis of <user feedback>.
After: Imagine you are preparing a report for the executive team. Provide a detailed analysis of <user feedback>.
Note: Some report that after offering a tip, ChatGPT generates longer answers. I confirm this. To see a difference, you need to use ChatGPT in different sessions. Otherwise, it will reuse the previous context of the conversation.
2.9 Iterate and provide feedback
In my opinion, that’s the most important tactic when working with AIs. If the initial response isn’t what you need, give constructive feedback and ask again.
Wrong feedback: Improve this product description.
Correct feedback: This product description needs to emphasize the unique selling points more. Focus on the benefits of the new features and make it more engaging. Also, simplify the language so that a primary school graduate can understand it.
2.10 Bonus: Reverse-engineer the prompt
Sometimes, the best way to create the right prompt is to provide AI with the expected answer and then reverse engineer the prompt.
Example: Act as an LLM expert. Help me create the prompt that might have resulted in the following outcome: <outcome> in the <context>. I will give you a $1,000 tip if you guess the right prompt.
3. Top 9 High-ROI LLM Use Cases for PMs
In the use cases below, I present:
Prompt template with variables
An example prompt
Results for the example
Note how much information (e.g., product, strategy, objectives, examples, templates, role) is required to get satisfying results. None of the presented prompts is a single sentence.
3.1 Came up with new ideas
Prompt template
You work in a product trio performing continuous product discovery for <product>.
Your goal is to <objective and desired outcomes>.
The product trio identified that <market segment, opportunity> wants to <desired outcomes> when <optional context>.
Ideate separately from the perspective of an experienced product manager, an experienced product designer, and an experienced software engineer. Come up with 5 product ideas for each role.
Prioritize ideas that <expectations>.
An example prompt
You work in a product trio performing continuous product discovery for an online trading platform (B2C, SaaS).
Your goal is to delight users with new onboarding, which will be measured by increasing the activation rate from 10% to 30%.
The product trio identified that inexperienced traders want to:
minimize the time it takes to decide how to invest their money,
maximize the likelihood of a positive return on investment.
Ideate separately from the perspective of an experienced product manager, an experienced product designer, and an experienced software engineer. Come up with 5 product ideas for each role.
Prioritize ideas that are most likely to solve the problem and are easy to implement.
Results for the example
Follow up questions
Describe <idea> in more detail so the product trio can understand it better.
Suggest <number> more ideas.
3.2 Identify hidden assumptions
Keep reading with a 7-day free trial
Subscribe to The Product Compass to keep reading this post and get 7 days of free access to the full post archives.