Everyone with a tech mandate is feeling pressure to use AI. We are told that AI can change the world, unlock hidden business value and solve problems that we weren’t able to before.
But how do you even start? What sort of considerations do you need to make? How do you even go about building your first AI product?
I faced these questions and more when I started thinking about launching my team’s first AI product. I didn’t have the depth of expertise in data science to grasp all the AI technical documentation out there and map them to potential use cases. The rest of the information I found was too high level. I needed to go beyond the hype around AI to define actual user scenarios and stories. There wasn’t a playbook to follow. We had to trail-blaze. It was hard.
There are a lot of materials that detail software product management best practices. But there is not a lot around AI product management. I spent some time documenting all the critical considerations that could make or break the AI product idea along the way in this article series. I created this article series for a tech-savvy, soon-to-be AI product manager who wants to know what the journey ahead will look like. My goal is for you to learn from our experiences so you can make the right decision at the right time along the way.
This article series was reviewed by Andrew Freed (Senior Technical Staff Member, Master Inventor and Data Scientist) who helps build and deliver AI solutions to clients and Jennifer Sukis (Design Principal for AI & Machine Learning) who is a Professor of Advanced Design for AI at the University of Texas. FYI, they’re brilliant and have a birds-eye view of what worked and didn’t work. These articles will soon become part of a playbook for other product teams inside IBM.
The 5 phases of the AI product management cycle below map to the high-level milestones of building an AI model, which follow a similar workflow. You’ll need to gather the training data, train the model, test the results, and repeat if you’re not getting the results you want.
The rest of the series will deep dive into each phase with a focus on product + business considerations over technical considerations.
The 5 Different Phases and Considerations
Phase 1: Go/No Go
Should I even use AI?
Are people ready?
Phase 2: Priority Setting
What goals and metrics should I use?
Phase 3: Training Data
How much data do I need?
Phase 4: Building the AI model
Who should be involved?
Phase 5: Testing and Deploying the AI model
When is the AI model good enough?
When am I done?
This workflow is very iterative and can take time to get right — you might get to Phase 5 and decide that it’s not good enough to release and then loop back into Phase 2 to Phase 4 again. You might release the product and then in Phase 5, gather feedback that your AI model isn’t performing as expected. This will bring you back to Phase 2 again. It’s helpful to think about AI product management as a “cycle” rather than a process with a concrete start and finish.
The articles will be published weekly — stay tuned to learn more about the 5 phases!
Stella Liu was a Product Manager at IBM Watson IoT where she helped build her team’s first AI-based product at scale. She loves to talk about AI, product management and environmental sustainability.
Please reach out to her at LinkedIn for any questions or comments!