Understanding the AI-Native Product

In this chapter, we will go over the essential components for creating a strategy for building an AI product. This strategy will allow companies a process that will help them succeed in building an AI native tool. In Chapter 2, we briefly introduced the new product development stages, and in this chapter, we will build on that structure by focusing on the most important phases of introducing a new AI/ML native product: ideation, data management, research, development, and deployment. We will also address the main contributors to your AI/ML product team, as well as the tech stack that will empower them. AI native products begin and end with your data, and the roles you fill to support the team responsible for their creation will be critical to their success. When building an AI native tool, do your due diligence so that you aren’t being wasteful of your company’s finances and resources. A bad hire or incorrect tech investment can be...

Stages of AI product development

Whether your organization is robust enough that you’re supporting multiple product teams for various stages of your AI product development or lean enough that you’re running on a skeleton crew that will see your product through each stage, you’ll want to be cognizant of what these different stages are so that you can define success through each phase. There are various schools of thought on what product management is or should be. We’ll do our best here to summarize the various phases in a way that best summarizes the core aspects of AI product development.

Either way, as an AI PM or leader of an AI product, you’ll want to consider how your product relates to each of the phases described in the following subsections so that you can identify the phase your product is currently in and what you need to do to bring it to full maturity.

Phase 1 – Ideation

Just as with the traditional software product life...

AI/ML product dream team

In this section, we will be spending some time understanding the various roles that will empower your AI product team to maximize success. We’ll also be grouping these functions across the stages we outlined in the preceding sections so that we can get a sense of when these roles come into play. Note that not all these roles will be necessary in your organization. Every organization is different and will have different needs. Use your discretion when building your AI teams. You may include other stakeholders in your AI program, but the following is a relatively complete list of the main stakeholders you will want to include in your hiring process as your AI/ML product team grows.

We’ll now look at a cumulative list of roles that will likely apply to your ideal AI dream team. We have listed the roles in order of common appearance.

AI PM

Here, we start on our journey of creating a dream team. All organizations are different. Some won’...

Investing in your tech stack

Understanding the tech stack and languages that will give you the most flexibility is the most important part of beginning your tech stack journey. In this phase, you’ll work closely with your data science and data engineering teams to create the proper channels for delivering the relevant data to your models in a reliable way so that all the other stakeholders involved in building your product can trust the infrastructure in place.

Managing ML experimentation is a formidable undertaking in and of itself, and we’ve seen tools such as MLflow and Weights & Biases used for managing versions and experiments. You can also use tools such as Cloudera Data Science Workbench, Seldon, Dataiku, DataRobot, Domino, SageMaker, and TensorFlow to support your data scientists with a workstation for building, experimenting with, deploying, and training ML models.

As a PM, you’re regularly thinking about the value of building something compared...

Productizing AI-powered outputs – how AI product management is different

In this section, we will be exploring the difference between product management for traditional products and product management for AI/ML products. At first glance, it may seem that AI/ML products aren’t that different from traditional products. We’re still creating a baseline of value, use, performance, and functionality and optimizing that baseline as best we can. This is true for every product as well as for the greater practice of product management, and this won’t change just because our product works with AI.

The true differentiator when it comes to AI products is you’re essentially productizing a service. Think about it for a moment. In order for AI to work, it has to learn from your (or your customer’s) data. Different models might work better on different kinds of data. Different datasets will require different hyperparameters from your models. This is the...

AI customization

Through the act of productizing, you’ll likely start to create groupings or cohorts for your product. You might find that certain models work best for certain kinds of customers and build a structure around that. The act of grouping your offering into use cases and communicating differently or optimally to those cohorts builds on the wisdom gained from productizing. Taking that a step further, you’ll then naturally start to verticalize.

Understanding the various considerations for business models, verticals, and special customer groupings will be an important part of how you go to market with your product. For instance, if you’re supporting a B2C or consumer product, you’ll want to invest more in information gathering by acquiring more of your end users’ direct feedback for your ideation phase. Because you’ll be creating a product that’s going to be experienced by potentially millions of users, you will want a strong...

Selling AI – product management as a higher octave 
of sales

With AI product management, you’re selling in a number of ways. There’s the traditional sense of selling, which is this: creating a product that your market wants to buy. This is inherent in any traditional PM role. Your understanding of your market, your product, and your salesforce is one where you’re confident in the solution you’re bringing to market, your solution’s performance, and your salesforce’s ability to articulate the value of the solution you’re building. Then, there’s the opportunity and challenge AI presents: the ability to sell the AI functionality itself.

Earlier in this book, we mentioned the difficulty AI/ML projects have in being deployed into production; this happens for a number of reasons, but among the top reasons is the inability to sell the value of AI to the broader organization. This will be incredibly relevant to any PM...

Summary

The work of a PM is never done. There are always more voices, perspectives, and considerations to take in. Coordinating all the stakeholders, technology, leadership, market analysis, customer feedback, and passion for a product isn’t an easy task. In this chapter, we covered the stages of the AI product development life cycle and the various roles that can make up your AI product dream team. We also covered the tech stack that can help that team build a product, and various focus areas to help that product stand out and resonate with the cohorts of groups that will be buying and using your product. Hopefully, this chapter has helped you understand what the most important factors are when you set out to build an AI native product.

As long as you’re hiring the right people for the roles you have open in your AI program, doing your due diligence to uncover the right strategy for tech stack adoption, structuring your product in a way that benefits your customers...

References

• The What Why and How of A/B Testing in Machine Learning: https://mlops.community/the-what-why-and-how-of-a-b-testing-in-ml/#:~:text=An%20A%2FB%20test%2C%20also,guesswork
• TFX: https://www.tensorflow.org/tfx
• Seldon: https://www.seldon.io/
• Dataiku: https://www.dataiku.com/
• DataRobot: https://www.datarobot.com/
• Domino Data Lab: https://www.dominodatalab.com/
• Cloudera Data Science Workbench: https://www.cloudera.com/products/data-science-and-engineering/data-science-workbench.html
• Weights and Biases: https://wandb.ai/site
• ML Flow: https://mlflow.org/