Creating Momentum Behind Your Machine Learning Project

Machine Learning is one of the most alluring tools among learning experience teams because it is at once accessible and intimidating. While the logic underpinning machine learning (ML) has aspects related to how we understand human learning, the statistics behind its execution can make any feasible pilot project seem like a pipe dream.

However, this stage of experimentation and pilots is where the vast majority of companies are at with AI at the start of 2020, and the missing piece to getting started is not necessarily learning the statistics and engineering behind it — though a foundational understanding is very helpful — but generating momentum behind your ML project.

A project using machine learning will likely require working with data scientists and data engineers, a highly sought-after group. To justify a collaboration, articulating the project’s potential to support a strategic priority and further data capabilities and culture throughout the business will form a strong case for support.

While this pitch work may seem a departure or distraction from learning product design and development, when you consider the type of insights ML offers, the value to learners of new and existing products is well worth the effort.

Having just begun a cross-functional ML project, I stumbled across four simple questions that forced me to clarify answers that ultimately resonated across teams and generated enough momentum to get a new project off the ground.

What’s Possible?

• What types of insights does ML supports?

What’s Useful?

• What types of scenarios and questions are fit for ML?

What’s the Goal?

• What’s the related business need and way forward?

What’s Needed?

• What will we need to get started and what minimal resources will we need to influence a culture of experimenting with data?

What’s Possible?

Before considering what opportunity to focus on, it’s helpful to know what machine learning is capable of. Simply put, ML learning is best used to rank, count, cluster or classify data. From a learner’s perspective, each of these functions could be valuably applied to support current or continued learning:

Classify: Apply something to a category.

• Popular Example: Classify an email as spam or classify a purchase as fraud.
• Learning Example: Curate all learning assets by type (medium, topic, time lengt, etc.).
• Learning Value: Simplifies the search for new learning experiences and provides data on usage and intent-based selection.

Count: Predict a numerical value of something.

• Popular Example: Predict car value by mileage.
• Learning Example: Show the market demand of a topic.
• Learning Value: Increases learner motivation by demonstrating professional relevance.

Rank: List items that are relevant to act on.

• Popular Example: Suggest products people with similar shopping carts purchased.
• Learning Example: Suggest the most relevant topic to learn next or recommend other learning experiences that were helpful to similar learners.
• Learning Value: Supports life-long learning and the beginner’s mindset. Tailor learning to an individual based on performance or satisfaction

Cluster: Group similar things together

• Popular Example: Show related news stories or all pictures with your best friend.
• Learning Example: Show other courses available on a general topic.
• Learning Value: Support self-directed learning.

These examples of how the core functions of ML can be applied are compelling not only to the learner, but the business as well. Consider how these same examples could grow or protect a business:

Rank:

• Learning Example: Suggest the most relevant things to learn next.
• Business Value: Retain customers.

Count:

• Learning Example: Show the market demand of learning a new topic .
• Business Value: Acquire customers who are undecided.

Cluster:

• Learning Example: Recommend other learning experiences that were helpful to similar learners.
• Business Value: Grow satisfied customers into brand promoters.

All of these features and more are possible with machine learning, but what’s useful? That will all depend on the different perspectives of stakeholders on the project:

• The data scientists and engineers you’ll collaborate with on the project.
• The business stakeholders who will fund/support your project.
• The learners who will experience the end result of your work.

What’s Useful?

The Data Scientists and Engineers

There may be no shortage of valuable machine learning projects to do at any one time; one way to increase likelihood of getting the green light is to generate interest from data scientists and engineers before pitching your project to gain support.

The simplest way to set up this collaboration for success is to gain credibility by demonstrating how the question and data underpinning your project is one where machine learning is useful and feasible.

Is it Useful?

Machine learning has a lot of cachet at the moment, but it is really helpful when it is tasked with providing insight to thoughtful questions. If you can find challenges and relevant data where ML is useful, you’ll be well equipped to demonstrate its applicability in providing insights.

Machine learning is good for challenges that:

• can’t be solved by a simple spreadsheet
• are constantly changing
• need to scale
• does not require perfect accuracy.

For any project, you’ll need to play with data. Machine learning is helpful when your data is:

• available — easy to pull
• sufficient — there’s enough of it
• unbiased — it’s reliable
• insightful — it’s relevant to your challenge
• ethical — it’s obtained with concern for privacy.

Of course, a good question with supporting data can show a way forward, but that has to be tied to the reality of your company’s capabilities.

Is it Feasible?

The machine learning capabilities vary by company, but if you’re trying to just get started with machine learning and show the value it has to your area of influence, one of the easiest ways to get started is with supervised learning, where you train the computer to learn using labelled data and example answers.

Of the core functions of ML — classify, count, rank, cluster — classifying and counting are best suited for supervised learning. This is where the proverbial low-hanging fruit resides. If you can develop a great regression or classification question, identify a few features and share a possible training data set, you’ll be in excellent shape to build an early relationship with your data science team because you’ve become a translator, a bridge between their team and other functional teams.

The Business Stakeholders

What’s useful to the business stakeholders funding your project is going to differ from what’s useful to the scientists and engineers you’re hoping to collaborate with.

As the hype around ML settles and many executives see it as another tool in the toolkit, the applications for an experiment are becoming clearer and simpler to frame, just as you might any cross-functional project needing support.

The tremendous potential of ML, however, differentiates it from other tools available, and you risk downplaying the upside at the risk of playing it cool. Through trial and error, I found these prompts guided me to a story that balanced the lofty and the pragmatic.

• Imagine a super power don’t we currently have. What does this look like?
• Consider how we could do that today. What is the goal of the project (i.e. what do we want to predict) and how does it give us this super power?
• How is this project feasible right now?
• How will we know if we’re successful?
• What will we be able to do if successful?
• How will learn quickly and what do we need to do it?

Many of these answers will likely come quick to you. The heart of this brief is…

What’s the Goal?

The folks supporting the project will likely want to know how it strategically aligns with critical business priorities. This can be easier said than done, but if priorities are not immediately clear, scan previous all-hands minutes, town hall agendas, or the reports from earning calls to glean where the opportunities are.

After that, consider how the central question to your project aligns with these near-term and long-term goals. If there’s not a direct connection, how is your project integral to the growth, stability, or quality of a core product? Maybe you’ve shed some light on a novel opportunity!

What’s Needed?

Assuming you’ve done some of the earlier homework to align your project to the business, identify relevant features and data sets, and create a relationship with your data team, figuring out much of what’s needed to execute the project can be guided by a data lead dedicated to this work.

There’s one last thing to consider, however, and that’s what the teams involved need in order to develop capabilities and culture that will be integral to continued success.

As consulting firms have suggested, the companies that are most successful with adopting ML at scale are those that invest equal parts in the human aspects of change — training, communications, community — as they do the technical aspects — tools, infrastructure, hiring.

With backgrounds in behavior change and cognition, learning teams are in a great position to communicate early on the support they need to develop capabilities around asking good questions and translating insights into action. Articulating this value early on in your work with ML gives an under-looked lever of change it’s due consideration. Adding this small ask at the end could have as large an impact on the overall company as what predictions stem from the project.

The wagon wheel is a common metaphor to describe an approach to supporting company wide transformation. There’s a central hub where you can pull from — be it a functional academy, lunch and learn series, data governance records, LMS, etc. — and there are spokes, the individual teams that develop new and existing capabilities. Consider what essential resources you need to update your team’s skills, processes, workflows, communications and other day-to-day aspects that need equal attention as the models and training sets you’ll use.

Along the way, document how this support impacted the outcomes of the project and what would be valuable from a central hub in the future.

In short…

there’s a lot of great things we can predict with machine learning. Finding an overlap between what’s possible, feasible, personally interesting, and exciting to the business is a trusty route to getting others fired up about your project.