Artificial intelligence (AI) isn’t magic. It also isn’t a product. It’s a feature. And like any feature, you should understand the problem, and the value proposition, before you write that first JIRA ticket. In this article, we’ll demystify machine learning for product managers and explore an MVP framework for product-led companies.
What Is Artificial Intelligence (Machine Learning)?
Artificial intelligence, a buzzword that is synonymous with machine learning, combines ideas from computer science, mathematics, and statistics to allow computer programs to generate insights and predictions without being explicitly programmed. Generally speaking, machine learning algorithms fall into one of three buckets: image analysis, text analysis, and numerical analysis.
Machine learning algorithms have applications in a variety of industries:
- Image analysis algorithms are leveraged by healthcare providers to identify and diagnose diseases using images and scans.
- Text analysis algorithms are popular among financial trading firms to gauge market sentiment.
- Numerical analysis algorithms are used by software companies looking to predict churn and measure customer satisfaction.
How Does Machine Learning Create Value for my Product?
Whether you realize it or not, machine learning is pervasive in products you use every day. Some popular examples include auto-correct/auto-complete in your email and phone, iPhone notifications on commute times and destinations, and recommendations on your Netflix homepage. The most magical part of these features is that very complex algorithms are wrapped into the simplest of product designs.
The most valuable (and powerful) machine learning use cases translate into hard ROI for product managers. Below are three examples of machine learning use cases that help drive revenue and productivity.
Recommendation Engines
Use cases: Converting customers from free trials to paid subscriptions, upsells, and customer retention
Examples:
- Clothing company Stitch Fix uses a recommendation algorithm to connect its clients to items tailored to their individual style preferences.
- Netflix’s recommendation system personalizes suggestions for which show or movie a user should watch next.
Computer Vision
Use cases: Productivity gains, smoother user experience
Examples:
- Intuit’s “SnapTax” product uses optical character recognization to allow users of their mobile app to take photos of their tax documents on their phone, then submit them for processing.
- Infinia ML applies their machine learning algorithms to ultrasound images to help medical professionals identify health problems.
Text Analysis
Use cases: Customer self-service, extracting insights from text
Examples:
- Bank of America’s “Erica” virtual assistant helps clients manage their finances through the company’s mobile app.
- Salesforce’s “Einstein AI” feature allows revenue teams to predict business outcomes, such as lifetime value and customer churn.
Given the rapid evolution of the machine learning field, novel use cases for ML come up every day. The bottom line for PM teams is that you shouldn’t go into ML for the sake of ML. Instead, pursue ML projects like you would any other feature and tie its results to KPIs that generate value for customers and your business. Remember: ML is a feature, not a product.
You Don’t Have to Be a “Machine Learning Expert” to Use Machine Learning
There are several open-source machine learning libraries that take care of the heavy lifting associated with algorithm development. Unless you are operating on the cutting or bleeding edge of your field, consider leveraging some of the most popular ML libraries used today: Tensorflow and PyTorch.
Some of the biggest advances in ML are only a few keystrokes away from transforming your product.
Machine Learning Strategies for PM Teams
The biggest difference between machine learning and traditional statistics is the data-hungry nature of its algorithms. Some of the top-performing ML models require thousands (or more) data points to generate good results. Furthermore, ML algorithms require a feedback loop to remain valuable. As a result, ML consumers must implicitly, or explicitly, validate model results so algorithms can learn and evolve over time.
As product teams identify new use cases for machine learning (or augment existing use cases), consider the following MVP framework:
- Do I even need machine learning?
- Sometimes, simple statistical techniques like linear/multiple regression can generate great results for your product. Machine learning may be overkill for your use case.
- What is the value for my user? How do I simplify the presentation of machine learning results into something that is consumable by anyone (e.g. think Netflix movie tiles)?
- What is the value for my product/business? How do I measure this value?
- Where does my “training” data (the base dataset you initially use to create an algorithm) live? How do I access it?
- What type of data is it (text, image, tabular)?
- Do I need to label the data in any way (e.g. positive review, negative review)?
- How do my algorithm’s predictions get validated? Where do I store this ‘validated’ data?
- Does it require an action from the user? (e.g. gives you a thumbs up or down)
- Does inaction indicate anything? (e.g. user ignores top five recommendations)
- How do I structure my data for future use?
- Consider what types of data you will be generating in the future and devise a strategy for automated data collection (and cleaning if necessary)
Expanding Your Strategy
The words “machine learning” and “artificial intelligence” can sound fairly daunting. As we’ve just seen, however, they are already part of our everyday lives. Don’t let ML or AI intimidate you or your team. Companies that embrace the opportunities presented by these tools create memorable customer experiences and offer value to users in new and unexpected ways.
