A practical guide to getting started with machine learning

Written by Pippa Armes  | 

5 min

 

Machine learning (ML) opens up a world of exciting possibilities for product managers. In a previous post, we talked about the ways ML is redefining the role of product manager—augmenting their capabilities and freeing up their time to focus on high-value tasks. 

But with the endless potential of ML and all the ways you could possibly apply it in your product, how do you know where to start?

When you can do literally anything, picking the one thing to do first can feel overwhelming. That’s why it’s important to build a foundational understanding of ML and ensure any projects you take on align to your customers’ needs, your company’s goals, and your product’s core purpose.

Hannah Chaplin (Director, Product Marketing at Pendo) and Dr. Inbal Budowski-Tal (Senior Director, Machine Learning at Pendo) sat down to discuss a few simple strategies to help project managers get started with ML during Product-Led Alliance’s AI for PMs Summit. Read on for the key takeaways from their session, or scroll to the bottom of the page to watch the full recording.

 

Where to start with new ML technologies

As you consider the range of possible ML use cases within your product, it’s helpful to think about where the technology could lend itself to the task most organically. Generally, areas of your product that involve analyzing a lot of data are a great fit for ML. Think processing large data sets, evaluating thousands of data points, or extrapolating themes from massive quantities of information—tasks that just wouldn’t be feasible for humans.

It can be tempting to start this process with a particular ML technology you want to try out in mind. But as the product manager, it’s important to remember that the problem—and your data science team—should actually be the ones dictating the technology you use.

When Pendo’s Machine Learning team decided to focus on Pendo Feedback as their first project, they started by evaluating the requirements they needed to meet. They found that they could potentially be achieved with two different ML tactics:

  • Supervised: a ML approach where you tell the algorithm what to look for in advance by providing it with a labeled data set
  • Unsupervised: a ML approach where you do not tell the algorithm anything in advance, leaving it to pull labels out based on the similarity of the objects

These approaches are fundamentally different from a data science perspective. But the decision to choose one over the other was ultimately based on product requirements and the desire to solve a recurring customer pain point: classifying large feedback data sets into categories for easier synthesis.

How to choose your first project

Dr. Budowski-Tal reiterated the importance of having a well-organized data collection process—something that came in handy as she was assembling Pendo’s Machine Learning group. She also noted that it’s crucial to lean on design partners for research and testing, and to help validate that the ML solution could deliver the best possible outcomes for all customers. 

Conduct interviews with customers and stakeholders across the organization—including product management, executive leadership, customer success, and engineering—to get a clear understanding of the challenges they see most often. From there, create a shortlist of ML project candidates, focusing on the commonalities between them that could potentially be solved by the same ML solution.

As you narrow down your list of project candidates, remember the “ME ME” rule: minimal effort, maximum effect. Find something small (but still impactful) to tackle first to help build trust with your customers and management team. Also remember that in the world of ML, even small tasks can take a long time. So it’s a good idea to keep your initial project relatively small—especially if it’s your first foray into ML. Draw clear boundaries around what your project will (and will not) address, and take a phased, iterative approach when rolling out your changes to production.

Building a business case—and the right team

Once you’ve identified the ML project you want to start with, you next need to assemble your working team and build a business case to present to leadership. Dr. Budowski-Tal noted that making the business case for ML is actually quite easy these days—because everyone wants it! But the more customer-centric you are throughout the entire ideation and validation process, the easier it will be to make a clear argument for your project and acquire the resources you need to get it done.

It’s also important to build a balanced team to actually work on your project. Start with a strong product manager to ground your team in customer-centricity and the ethos of your product. You’ll also need a group of data scientists and machine learning engineers to identify the right ML solutions, and back and front-end engineers to build them. Finally, a familiar project management framework like Scrum can be incredibly helpful for keeping everyone organized and on-task.

A few more tips

Here are a few more helpful tips for product managers looking to get started in ML:

  • Always put the impact before the science—focus on the customer problem and let the technology follow
  • Be curious and have a data-informed mindset
  • Get comfortable with being uncomfortable—by nature, there’s risk and uncertainty involved in ML
  • Partner closely with your ML team and ask them to help you learn as you go

 

Want to learn more about machine learning? Watch the full recording from the Product-Led Alliance AI for PMs Summit here: