AI/ML Product Management Fundamentals

How do you build products that leverage machine learning?

Machine learning is using data to answer valuable business questions.

Answering these business questions should lead to the creation of tangible business value. This could be increased revenue, decreased costs, increased retention rate, increased operational efficiency, etc.

Therefore, always focus on the business impact of artificial intelligence when building AI/ML products.

You also have to understand when to use AI/ML, as it may not always be the right solution for a given business problem.

Before diving into building and training machine learning models, there are strategic elements that we must explore beforehand. The success of an AI/ML product depends on getting the strategy right.

Here are the 3 building blocks of AI/ML product management:

1) Business problem and market

AI/ML product management always begins with identifying important business problems. In order to identify business problems in the first place, you must have a strong understanding of the business domain or industry.

In my article titled, "The Most Important Element Of AI Adoption", we cover the end-to-end machine learning process, and why domain knowledge of the specific business/industry is paramount for successful AI adoption.

Start out by identifying 3-5 candidate business problems, and then assess whether AI/ML is the right solution to solve them (this topic can make its own, long article).

For the business problems that made it through the previous step, the next step is to define clearly who is having these problems. This is what we call the target market for our AI product(s).

In my YouTube video titled, "How To Identify AI/ML Use Cases" we covered the 3 different types of AI/ML product use cases: Internal, B2C, and B2B.

If you are building an internal machine learning platform, the target market may be internal data scientists, data engineers, and/or machine learning engineers. An AI/ML product manager for this type of product would require a more technical background given the technical target market.

If you are building a B2C mobile app that leverages computer vision to help identify the most common skin diseases, your target market may be a specific consumer segment. An AI/ML product manager for this type of product would require a strong understanding of the healthcare domain, ideally dermatology.

If you are building a B2B product that helps organizations label unstructured data quickly and efficiently, you target market may be AI product managers. An AI/ML product manager for this type of product would require a strong understanding of the intricacies and impact of datasets for deep learning use cases.

It goes without saying that any type of AI product manager requires a strong understanding of the end-to-end machine learning process.

It's essential to understand your target market's needs, pains, and desires. AI product development is driven by customer development, and we must prioritize accordingly (we will cover prioritization of AI adoption and AI product development in a future article).

Once we have identified a business problem, framed it as a machine learning problem, and identified the target market, the next step is to assess dataset requirements.

2) Dataset scoping and refinement

This is where we begin to get into AI/ML product design.

However, AI product design is different than traditional software product design.

As the AI legend Andrew Ng said: "Datasets are the new wireframes."

Therefore, instead of sketching app screens, begin by scoping out a dataset prior to building/hacking an MVP.

In supervised learning, a dataset is a collection of features (data) and corresponding labels (the answers we want to predict based on the patterns within the data).

Here are some questions to ask when defining your data strategy:

• What are the features that encapsulate the prediction problem? (Requires domain knowledge)
• What are the specific labels we need? (It's not always straightforward, even in supervised classification)
• Precisely, what data should we collect? Do we have it already? Is it high quality and representative of the problem we are trying to solve?
• How should we label/annotate this data? Who should do it, and what specific instructions do they need to minimize labeling errors?
• What are the edge cases?
• How do we minimize bias?

Datasets make or break the performance of machine learning models.

Great data >> great models

The predictive power of your machine learning models is directly proportional to the quality and quantity of data you used to train them.

Make sure you invest enough time upfront on your data strategy. Also, refine datasets over time and be open to iteration as you monitor user behavior and your chosen metrics.

3) Optimizing metrics and iterating

Once we understand the business problem, target market, and dataset requirements, it's time to choose actionable metrics to measure the effectiveness of our solution.

You have to choose the correct metrics based on the business problem you are solving. There are two kinds of metrics for AI/ML products:

• Business outcomes
• Model outputs

Business outcomes are the most important metrics to track and improve. Here are some questions you can ask to help you define metrics that involve tangible business value:

• Is the product improving the company’s bottom line? How much?
• Is it helping generate revenue? How much?
• Is it improving operational efficiency? How do we quantify it?
• Is it contributing to forms of intangible business value, such as improving customer satisfaction?

Define metrics for business outcomes precisely and quantitatively. Track them continuously so you always know whether you are getting closer to your goals or not. If not, adjust/iterate accordingly.

Model outputs are the predictions made by the machine learning models. Since ML models cannot be 100% accurate, build a process into the product itself to handle wrong predictions, see the data where the model is making mistakes, and use this feedback to improve the dataset accordingly.

Also, make sure wrong ML model predictions are handled gently within the product to maximize the user experience.

As an AI/ML product manager, always make decisions based on data and metrics.

Check out my article titled, "AI/ML Product Success Metrics" to explore business outcomes and model outputs in more detail, including real use cases.

You may have noticed that we did not go into more technical details, such as machine learning algorithms, neural networks, or any mention of coding.

This is because AI/ML product managers are not involved in day-to-day coding. Yes, they are required to have a strong understanding of technical concepts in ML, but it's not their job to write the code themselves. Other members of our team do that, such as data scientists, ML engineers, and software engineers.

That's why this career path is attractive to many people. They love artificial intelligence, they love business strategy, tech products, solving business problems, and leading execution to achieve business objectives. However, coding is not necessarily their strong suit or preference.

AI/ML product management is a highly rewarding and challenging role that will continue to grow over time as more companies adopt artificial intelligence into their products.

If you need help to accelerate your company's machine learning efforts, or if you need help getting started with enterprise AI adoption, send me a LinkedIn message or email me at [email protected] and I will be happy to help you.

Subscribe to my blog to get the latest tactics and strategies to thrive in this new era of AI and machine learning.

Subscribe to my YouTube channel for business AI video tutorials and technical hands-on tutorials.

Client case studies and testimonials: https://carloslaraai.com/enterprise-case-studies/

Follow me for more content: linkedin.com/in/CarlosLaraAI

#ai #artificialintelligence #machinelearning #deeplearning #leadership #aiadoption #productmanagement #aiproducts #aiproductmanagement