Product Manager’s Guide to Leveraging Generative AI: A 3-Step Guide
In a world surrounded with emerging tech, how can we create the next big thing using AI?
In a recent artificial intelligence workshop I attended, attendees were asked to create a product.
Create a 1-minute informational video teaching me any topic of your choice. You can use any tools you want. You have one hour.
While many focused on deliberately crafting a script or researching engaging information, I took a different approach: Using every AI tool I knew.
In one hour, I was able to produce a dense script, film a video, generate captions, and add supplemental visuals, all with the help of AI.
The disparity of our output efficiency was eye-opening. As an aspiring Product Manager, I was able to optimize the development process with the assistance of generative AI. Here are the top three AI tools I used and how they can superpower your product management workflow.
We can divide the Product Management process into three general steps: Market Research (analyzing what consumers want), Product Prototyping (building the product), and Product Development (finding consumers and investors). For each step, I’ll provide a brief description of what happens in the process and an AI tool to upgrade a PM’s workflow. However, I’ve always preached that a tool is only as good as the person using it, so for each tool, I’ll not only go over a brief overview of how it works, but a tip to leverage it to our advantage.
Step 1: Market Research (ChatGPT…Kinda)
After formulating a product idea, Product Managers need to find out if there is enough demand for a product. The best way to do this is to see what has already been working to see what the market demands. For example, if we want to build a mobile app, we have to analyze what other companies have done the same thing, what ratings they receive, how many downloads occur daily, what average age of consumers access similar apps, and other common data points. However, this means a lot of data needs to be analyzed. This is where a not so known feature of ChatGPT comes in: Data Science Analysis.
Most of us know ChatGPT to generate essays or to summarize some short prompts, but this is only the tip of the iceberg. Reddit user hjras created a visual to explain this, with tasks like providing quick summaries of text at the top of the iceberg, generating project ideas in the middle, and predicting future trends based on data on the very bottom. Did I mention that the visual was created by ChatGPT?
But how does ChatGPT analyze data?
In order to analyze data, ChatGPT can import Python libraries, similar to how we can import API’s or software into our Python environments. The two most common are Seaborn and Pandas, two libraries used to analyze data.
These python libraries “read” datasets by converting data into bytes. We can use an analogy of moving from one house to another to understand this concept. If we need to move all of our items, we could technically package each of them individually, but this would be inefficient when we start scaling the magnitude of objects to package. Instead, we can organize all the objects into standardized boxes. This way, each box is designed to hold various types of items efficiently and is easier to manage. In computer science terms, each item in our house, or dataset, is a data point. Instead of moving each item or analyzing each data point one by one, the data is organized into groups to be analyzed more effeciently. This way, if we want to take stock of what types of furniture we have, we can go straight to that box instead of rummaging through random items until we find what we want.
So how can we utilize this?
Since Large Language Models (LLMs) like ChatGPT are so effecient at analyzing large datasets, go crazy! When we want to train any type of model, the more data we have, the more accurate predictions will be. Even if you don’t feel like uploading a spreadsheet and dealing with plugins, ChatGPT-4 can handle up to 25,000 words at a time. Going back to our app example, we can ask the language model to ingest thousands of statistics about our app, and use it as a consultant to answer questions like “How many downloads do the top apps in our niche receive on average?” or “Should we create targeted marketing campaigns for specific demographics or regions?”. The possibilities are (almost) endless.
Step 2: Product Prototyping/Engineering (Figma Plugins)
Now that we know that the demand for a product is there, we need to create a product prototype to really visualize what we are building. Usually, this consists of visualizing the product in design tools like Figma before coding it into a working prototype using HTML, CSS, or other coding languages. However, while Figma’s easy learning curve allows PM’s creative imagination to run free, coding is not so easy. While individuals may have the creative ability to think of successful products, their technical skills throttle their ability to share it with the world…until now.
Realizing this problem, Figma designers created a plugin (software add-ons to enhance capabilities) to bring these designs to life: Figma to HTML.
But how does this work?
In order to analyze an image, AI programs can scan Figma frames using machine learning and computer vision. But how does a computer learn to recognize what is a button versus an image? In order to learn the difference, the computer program must be trained with a variety of different buttons and images. Each time the computer is given an input, it will try to guess what it is seeing and then receive feedback based on how accurate its response was. This way, the computer can learn from its mistakes and become more and more accurate the more it is exposed to different inputs.
After a computer is trained, it can now recognize a variety of elements on our website like titles, images, and buttons. Then, it creates what is called a HTML Document Object Model (DOM). This organizes the features on the website almost chronologically so it can organize your Figma frame into a chronollogical format from top to bottom.
So how can we utilize this?
While it isn’t as easy to take advantage of the technical aspect of this tool like some of the others, learning its use case can greatly help accelerate the process. Instead of spending your time learning how to create an article using <div> tags, spend your time learning how to create a beautiful project. In the future, there will be thousands of AI tools able to generate code, but we are the ones who need to visualize and communicate the idea. Spend your time researching digital styles that work for you, and let plugins like Figma to HTML handle all the technical work at the click of a button.
Step 3: Product Development (Gamma)
Finally! We have our product and are ready to share it with the world! However, in order to create a prospering product cycle, we need to collect investors and attract consumers to our product. For this, we can use a tool called Gamma. Gamma is a website used for creating slidedecks, presentations, or websites based on a user’s prompt.
But how does this work?
AI generation programs use what is called Natural Language Processsing (NLP) to understand and respond to human language. NLP’s look at key words in a request and use that to generate an output, one that is hopefully accurate to the user’s idea.
For example, let’s say I am trying to describe how to create a website with the prompt: “Create me a website used as a database for sports equipment. Each piece of equipment should have a rating system up to 5 Stars.”
The NLP will then recognize key words: “Create me a website used as a database for sports equipment. Each piece of equipment should have a rating system up to 5 Stars.” Based on these words, the program knows what the user wants and is able to send it to a Generative Adversarial Network (GAN) which is a machine learning model used to generate website designs and images for those websites.
So how can we utilize this?
Because we know that NLP’s look for key words, make your prompts as specific as possible! These programs scan your prompt for every feature that you want added, so the more you describe about your website the better. By doing this, you save more time describing what we want to the AI program instead of manually inserting features later on. Additionally, use these quick generation tools to visualize what your final website or deck will look like. We’ve all heard of writer’s block, the feeling where we don’t know what to write or how to start which leaves us pondering for hours on end. However, tools don’t neccesarily generate our finished product, but can give us the guideline and format to accelerate our workflow that extra step quicker.
So what can we conclude?
Besides the fact that there are a multitude of AI tools for every step of the process, the technicalities behind each tool is what really matters. From each tool, we noticed that each program is optimized for handling large inputs of information. From data analytics to slide deck generation, AI tools can handle loads of information, so load it to the brim to get the best results possible.
While my video created with these might have scratched the surface of AI’s full potential, there’s one last thing to note. Behind all of the unique tools and groundbreaking technology mentioned above is a Product Manager who has the ability to use them effectively. We need to make sure we are not only familiar with different tools, but are able to use them to their full potential. While the tools seem elusive and magical, the next step forward will be determining how well a Product Manager can wield them to their advantage…
