Navigating AI Product Management: My Journey Building an AI Chatbot Prototype

TL;DR

I recently completed the Complete AI Product Leadership Blueprint course to elevate my AI product management skills and earn a certification to set myself apart in this rapidly evolving field. The course capstone project was to create an AI prototype primarily using no-code/low-code tools.

I chose to develop an interactive AI chatbot that combines a research e-commerce transaction dataset with user feedback to create a product recommendation engine. This engine immediately incorporates user feedback into the algorithm that determines what to display next.

Since this prototype doesn’t connect with any real e-commerce purchase flows, I named it the “Popular Products Showcase.”

I hope you’ll try the Popular Products Showcase AI chatbot and share your feedback using the web form below the chatbot. Links to my chatbot prototype and the AI Product Leadership course I took are at the end of this post.

What Makes This Chatbot AI?

The Popular Products Showcase is a conversational AI chatbot experience. Users are presented with products based on historical e-commerce transaction data, and their “Likes” are recorded through the prototype.

“Likes” influence the popularity ranking of products. Therefore, user input directly and immediately impacts the display algorithm.

It’s simple—you can easily see how the algorithm learns from your actions and adapts what it displays next.

Data Sources

I used a research dataset from Kaggle.com as the basis for the core transaction data.

To create a visually engaging experience, I used DALL-E to generate AI-produced product images, as the e-commerce transaction dataset did not include any. Some of these images are beautifully rendered and make sense based on the product names, categories, and brands. Others are more playful, artistic interpretations.

Tool Selection and Evaluation

Choosing the right tools was the first challenge.

? Front-End Experience: I explored Voiceflow and Bubble.io, ultimately selecting Voiceflow for its simple end-user experience, integration capabilities, and user-friendly chatbot development interface. This choice allowed me to focus on the recommendation engine and data flow.

? Database Management: I started with Google Sheets but transitioned to Airtable, recognizing the need for more advanced functionality. Airtable provided robust features and integration options, which were invaluable in managing the large volume of transaction data. Its stronger capabilities for posting and updating individual rows and fields were particularly beneficial.

? AI Models: I evaluated ChatGPT 3.5 and 4o, selecting ChatGPT 4o for its superior conversational flow and name resolution. This ensured natural interactions and a better user experience.

Development and Iteration

With the tools selected, I narrowed the project scope to focus on delivering one core use case successfully before adding additional features.

I implemented a deterministic time filter with preset options, allowing users to select from “Today,” “Yesterday,” “Last Week,” and “Last Month.” This simplification reduced the workflow complexity.

To enhance visual engagement, I initially created DALL-E images for the top 200 products based on frequency in the e-commerce transaction dataset, using placeholders for the rest. I created images for the remaining product images later.

To optimize performance, I reduced the dataset from 50,000 to 12,000 rows. This smaller sample set of transactions per day enables users to see how their “Likes” immediately impact the recommendation algorithm, making the experience more interactive and engaging.

Challenges and Solutions

During development, I encountered several challenges, including:

? Limited Dataset: The dataset covered only seven months of transactions. To address this, I used AI tools to create synthetic transactions, extending the data to a full year. This ensured the chatbot would continue providing varied user experiences over time.

? Airtable Limitations: Airtable’s lack of a UNIQUE function (available in Excel or Google Sheets) required restructuring my data tables to compensate.

? CSV Export Issues: Airtable’s lack of a native CSV export function forced reliance on third-party plugins. While this could have introduced potential PII exposure risks, the dataset I used was cleansed and fictitious, containing no actual PII.

? Subscription Tiers: Nuanced functionality differences between free and paid tiers in both Voiceflow and Airtable were not clearly communicated. I ultimately subscribed to paid tiers for both services to access the features I needed.

Reflections

Here are the key lessons I learned during this journey:

1. Plan Early: Early planning of user workflows and data flows is essential to avoid significant rework later.

2. Evaluate Thoroughly: Tool limitations highlighted the importance of thorough evaluations. Clearer communication by tool providers regarding tier-specific features would be greatly beneficial.

3. Empathize with Engineers: Debugging issues was complex and time-consuming. This experience deepened my appreciation for engineering teams and their work.

Future Directions On This Prototype

This prototype is a v1 product, and I see significant opportunities for improvement. My top five priorities for future iterations are:

1. Enhanced Recommendations: Leverage collective and individual user data to refine AI-based recommendations (e.g., personalized suggestions based solely on a user’s preferences).

2. Flexible Time Filters: Move from four fixed timeframes to customizable ones.

3. Limit Like Activity: Introduce a one-like-per-product-per-day rule to reduce the potential for algorithm manipulation.

4. User-Initiated Searches: Enable users to search by product brand, category, price, or any other metadata.

5. Incorporate Profit Margins: Factor in profit margins to recommend more profitable products in tie-break situations, benefiting business outcomes.

The Broader View on the Impact of AI Tools

This AI Product Leadership Blueprint course and its capstone project have been an enjoyable and valuable learning experience.

As an AI optimist, I’m continually inspired by the increasing opportunities available to all of us from the ever growing set of AI tools used by product managers, designers, and engineers. Companies embracing this revolution, investing in technology, and training their teams will lead the next wave of creating transformative products and services for customers, driving outsized value for investors and shareholders.

However, it’s crucial not to lose focus on the human aspect of this change. Supporting cultural transformation and leading teams through uncertainty will be essential to turn fear into excitement, drive adoption, and accelerate value creation.

I invite you to explore my chatbot prototype and learn more about the AI Product Leadership Blueprint course. Your feedback is greatly appreciated!