Making the Business Case for Machine Learning: Advice from the Experts

On October 24, the Next Matters Most MeetUp held a panel discussion on Machine Learning. The event brought Machine Learning experts together with members of the Triangle Community who (like me) wanted to learn more.

While people know Machine Learning is out there and can explain the tech value of it, the current challenge is how to demonstrate the business value. The panelists shared different ways of doing this effectively — and we even got some great ROI stats courtesy of Richard Boyd.

The conversation was non-stop throughout the evening with lots of “war stories” and cautionary tales shared. It was great to see people arrive early and stay for the 30+ minutes of Q&A after the panel ended.

For those who couldn’t attend, we will be sharing a comprehensive recap of the evening over a few posts. If you’d like to watch the panel yourself, we live streamed the panel on Periscope and the recording is available on our Twitter feed.

Quick Intro to the Panelists

The panel featured five really smart speakers who have been using ML to bring value to their organizations for years.

• Moderator Ken Wood is a machine learning practitioner and founder of Roar Marketing Concepts LLC, a data-driven marketing strategy consultancy.
• Serial Entrepreneur Richard Boyd is CEO and co-founder of AI and machine learning company Tanjo, Inc. and simulation learning company Ultisim Inc.
• Eric Reifsnider, Ph.D. is Data Science Manager at Validic, the leading platform for patient-generated health data integration and analysis.
• Brooks Adcock is Director of Innovation at operations management software provider Dude Solutions.
• Dawn Code is co-founder of Unspecified, LLC, which provides higher quality software and specializes in machine learning.

Their full bios are available here.

Making the Business Case

Ken got the evening started by asking the panel to share what value business owners should

anticipate from Machine Learning.

Richard believes the business case for Machine Learning is starting to take shape. Companies he has worked with are seeing 10x ROI on projects taking six months to implement and costing around $200k (or somewhere in the six figure range). He even mentioned having several Fortune 100 companies get 1500x ROI.

At the end of the day, Machine Learning brings measurable ROI plus annuity and this makes it easy to sell the value up the chain.

Brooks took a more philosophical approach pointing out two main business drivers for Machine Learning: scalability and forecasting.

Machine Learning enables teams to scale efforts and continue to perform at a high level without increasing headcount. He shared a great example of how a service organization could automate tasks (especially routine or mundane ones) and free reps up to interact with more clients.

Brooks also credits Machine Learning as leading to better and quicker decision making. This type of optimized forecasting, which is easy to quantify, gives you the edge over competitors.

Machine Learning: When It Makes Sense

In certain situations, Machine Learning is a better solution and can solve business problems.

For Dawn, if you are looking for patterns then Machine Learning is the way to go.

But Eric cautioned that it is not a magic bullet. Machine Learning doesn’t make things possible — it simply makes things much more practical.

When it comes to use cases, Richard advises big organizations (such as institutions of higher education, consumer goods companies and manufacturers) to look at every activity being done internally and determine what humans should be doing and what should you consider turning over to machine attention and effort.

— — -

What You Need to Succeed: Skills, Resources & Team Members

Next, Ken asked how business owners could be successful with Machine Learning. What resources and skills are needed? This question yielded a lot of great advice from the panelists that came as a result of several cautionary tales.

For Richard, executive buy-in for the project is a must. People are threatened by ML and often fail to realize how it actually benefits them (frees them up to concentrate on high value tasks by automating boring work, etc.). To be successful, you need to have top cover that understands how ML will benefit the company and to push through barriers.

Eric recommends appointing someone to “own” ML internally and act as the interface with any outside providers. Ideally, this person has some existing understanding of ML but they only need to be able to “talk the talk” (not do the heavy lifting).

Brooks shared a detailed blueprint for success. First, start with a narrowly-scoped, practical business strategy and then put together an internal “dream team comprised of:

• One project owner with executive backing to spearhead the project
• Strong DevOps person to securely penetrate internal data silos while keeping CSO happy
• Data scientist to correct missing info and get to the point of millions of records
• Good product person to build externally facing system with the data for end consumers to use

Lastly, have a centralized infrastructure like AWS as the foundation of the project.

Cost of Getting It Wrong

One concern that Ken shared is the fear of getting it wrong. Business owners worry about the cost of errors (what if something is classified incorrectly or a wrong decision is made) — this thinking can derail ML projects.

The panel agreed that the ROI in Machine Learning is so strong that you don’t need to get it right all the time for it to still be extremely beneficial, and that the chances that human trained machines have a stronger chance of getting it right than an individual human themself.

Brooks advises using the following calculation to ensure you are still coming out ahead:

Take the # of instances that were wrong and multiply them by the cost of the errors. As long as your overall cost savings exceed this figure — you are still making money with ML.

Challenges to Anticipate: Data Issues & Questions

When it comes to what business owners should expect in the way of obstacles the entire panel agreed that data issues are a given.

Machine Learning brings a reality check to your data. It sheds a light on bad and missing data and identifies data entry issues. Sometimes it can give you a a bit of a wake-up call. However, there is value in learning what you actually have. Just be ready for some hard conversations around rectifying data issues and enacting a sound data governance plan.

Dawn regards ML as a platform that actually creates more questions. It allows you to explore what is really going on with your data all at once (in a way humans can not). What knowledge is gleaned from Machine Learning? What does it mean?

According to Dawn the answer varies: it could mean nothing or it could point you in a completely different direction. So be open to questions (not just answers) when you get started.

Learning 101: How to Get Educated
Machine Learning is an extremely broad field. When it comes to learning and education — Eric shared some of the night’s best advice.

Machine Learning is the solution to a problem and there are lots of different problems to solve. Focus on problems you care about. Identify a specific problem you want Machine Learning to solve and then work backwards to determine exactly what you need to learn.

This is solid advice and gives you a concrete way to tinker and apply your knowledge.

Some of the tools and courses recommended by panel includes:

• Online courses from Udacity and Stanford University
• Python — its free, mature and powerful
• OpenAI: non-profit AI research company

I am amazed that all of that info was shared in a 45-minute panel discussion.

I would like to extend a huge thank you to Ken, Richard, Eric, Brooks and Dawn for sharing their time and expertise.

It was great to meet so many curious, like-minded people that evening. I know everyone who attended came away with a better understanding of the business impact of ML and the different strategies needed to be successful.

For more conversations like this, head on over to www.nextmattersmost.com

Interested in learning even more? We’ll be holding a workshop on UX For Machine Learning on November 28th at 6pm. Email hello@smashingboxes.com for more info.