Productizing the ML Service

In Chapter 6, we briefly touched upon the notion of productizing and what that means for AI outputs in the Productizing AI-powered outputs – how AI product management is different section. We will be expanding on that concept in this chapter by exploring the trials and tribulations that may come up when building an AI product. Rather than thinking of AI products as traditional software products, it helps to think of them as a service that you’re learning to productize. What this refers to is the ability to create a consistent workflow that you can rely on to deliver consistent results in the way traditional products demand.

We will be going more in depth into product management principles and aligning them to the idiosyncrasies of AI/ML services.

By the end of this chapter, we will have an understanding of the following topics:

• Understanding the differences between AI and traditional software products
• B2B versus B2C –...

Understanding the differences between AI and traditional software products

There are a number of differences between traditional software products and AI/ML products. In the following subsections, we’ll first go over how they’re similar, and then we’ll note the differences between the two to give us a well-rounded sense of what to expect when you’re product managing an AI/ML product. This will help us establish a baseline as well as a deviation from traditional PM work. When you’re a PM, you’re often tasked with being the person to maintain an intuition about your product and how it will grow and evolve through the process of building and shipping the product and working with your engineering team.

Part of that intuition will relate to how you will market and sell your product, what kinds of customer needs and issues your product can anticipate, as well as potential problems that might arise as you start to get into the weeds with building...

B2B versus B2C – productizing business models

When it comes to building and shipping products, some of the biggest differences between B2B and B2C business models include domain knowledge and the degree of experimentation. In this section, we’ll be focusing on those two areas because they have the biggest impact on what productizing looks like for AI/ML products between these two business models. If we expand on the notion that AI/ML products behave more like services, the desired end result of both these business models will be different because they serve different kinds of customers and different overall needs.

With B2B products, there’s a strong need for these products to demonstrate a high degree of domain knowledge and a focus on that. Since B2B products are often serving a proven business niche, they must often prove they have expertise in this niche and have studied it thoroughly enough to be able to deliver on a need. With B2C products, we see a focus...

Consistency and AIOps/MLOps – reliance and trust

Maintaining trust, reliance, and consistency within your internal product teams as well as with your customer base is an act of committing to a specific ritual. Ritualizing the acquisition of clean data, tracking the flow through your infrastructure, tracking your model training, versions, and experiments, setting up a deployment schedule, and monitoring pipelines that get pushed to production are all part of the necessary work that needs to be done to make sure there’s a handle on the comings and goings of your AI/ML pipeline. This ritualizing is what’s referred to as MLOps or AIOps. In this section, we will explore the benefits of AIOps/MLOps and how they help you stay consistent.

If you’re managing an ML pipeline, you will need to learn how to depend on an MLOps team and set up your team for success. You don’t want to get caught losing $20,000 in 10 minutes (as we saw in our Profit margins section...

Performance evaluation – testing, retraining, and hyperparameter tuning

MLOps helps us with accentuating the importance of retraining and hyperparameter tuning our models to deliver performance. Without having a built-out AI/ML pipeline that validates, trains, and retrains regularly, you won’t have a great handle on your product’s performance. Your MLOps team will essentially be made up of data scientists and ML and DL engineers that will be tasked with making adjustments to the hyperparameters of your model builds, testing those models, and retraining them when needed. This will need to be done in conjunction with managing the data needed to feed this testing, along with the code base for your product’s interface as well.

In addition to testing and validating the models and working to clean and explore the data, MLOps team members also traditionally do software testing such as code tests, unit testing, and integration testing. In many cases, your AI...

Feedback loop – relationship building

Continuously monitoring and reinforcing the legitimacy of a complicated system such as an AI/ML pipeline is all in service of the ultimate goal of building relationships that last. Relationships between your company and customers, your development team and your sales team, and your MLOps team and your leadership team are all forged through this work of building and going to market with your AI/ML native product. In AI/ML products, the feedback loop is everything. Nurturing a strong relationship with the builders of these products and the customers they serve is the underlying work of the PM. Productizing is the process of taking a service, process, skill, or idea and finding a way to present that to the greater market. Many layers of work go into accomplishing this well, but at its most basic level, this work is really just an elaborate feedback loop.

We haven’t discussed marketing much in this chapter, but this will also be an...

Summary

The act of productizing involves taking a concept, a service, or a piece of technology and developing it into a commercial product that’s suitable for the customers you’re looking to attract. As we’ve seen throughout this chapter, this work isn’t just a matter of getting your product up and running and creating a landing page for your potential customers to magically find. Productizing involves critically understanding the business model you’re working in and the ultimate audience you’re building for. Remember that AI products can be thought of as AI/ML services that are being built into traditional software products. This means that another big part of productizing for AI products involves the standardization and ritualization of the AI/ML service in a way that’s repeatable and predictable for the internal operations teams as well as the customers that will come to rely on your product.

If you’re able to understand...

References

• Past The AI Hype: When Is AI Actually Profitable?: https://www.forbes.com/sites/forbesfinancecouncil/2020/09/01/past-the-ai-hype-when-is-ai-actually-profitable/?sh=3fdab9b81fb4
• How to Scale AI in Your Organization: https://hbr.org/2022/03/how-to-scale-ai-in-your-organization
• What are the net profit margins of a SaaS company/startup?: https://onplan.co/blog/what-are-the-net-profit-margins-of-a-saas-company-startup/#:~:text=Based%20on%20a%20KeyBank%20Capital,between%2068%25%20and%2075%25
• Guide To Data Cleaning: Definition, Benefits, Components, And How To Clean Your Data: https://www.tableau.com/learn/articles/what-is-data-cleaning#:~:text=tools%20and%20software-,What%20is%20data%20cleaning%3F,to%20be%20duplicated%20or%20mislabeled
• Why Commercial Artificial Intelligence Products Do Not Scale: https://ai-forum.com/opinion/why-commercial-artificial-intelligence-products-do-not-scale/