3 ways to make machine learning in business more effective

Decide what you want to improve

Being very clear about how you would like machine learning to improve your business might seem like a no-brainer, but you'd be surprised how many companies want to adopt machine learning simply because it's cutting-edge technology -- not because they have a plan to use it to solve any of their real business challenges.

The first step toward successfully embracing machine learning as an instrument of business improvement is to clearly define what you want to do better. At Dun & Bradstreet, we know what we want from our work with machine learning: We want to help our customers make smarter business decisions based on better predictive insights using our data and analytics.

For example, we are iteratively experimenting with machine learning-based risk analytics. Machine learning algorithms with adaptive components are used to create fraud scores that help address the constantly changing behavior of fraudsters and to analyze credit risks.

"In applications such as customer delinquency models, failure score, and fraud risk index, machine learning methods have already provided immense value and complemented existing methodologies," said Alla Kramskaia, global leader of advanced analytics at Dun & Bradstreet, in a May 2018 article posted on our website.

You also want to use machine learning in business in different ways for different tasks. If you have a task where it's acceptable to simply do a satisfactory job of delivering basic insights, then machine learning without much human input should serve your needs. For example, an algorithm built to provide automated news alerts about suppliers can notify you on its own if a company in your supply chain is filing for bankruptcy.

However, if the task is more complex and can help you demonstrate or even sharpen your organization's competitive edge, just running machine learning software won't be sufficient to outperform competitors. For complex, high-stakes business tasks, coupling machine learning with human insight is highly recommended. In fact, that's my next recommendation.

Use your people to inform machine learning

Including human intelligence as part of the process of complex machine learning applications really is non-negotiable. Even though analytics technology is improving rapidly, the human mind is still the gold standard for creativity and complex ideation. There's huge potential for machine learning in business, but that potential can be deceptive for analytics teams that want to use the technology to automate everything without regard for how doing so can affect the quality of the analytics output.

"Some practitioners feel that once ML and AI are enabled, a simple press of a button solves all problems, which is false," said Karolina Kierzkowski, leader and senior director of global data science at Dun & Bradstreet. Machine learning tools "are enablers for us to scale and create bigger and greater capabilities" for analytics, she said.

But to drive deeper analytical insights and business value, these tools need to be put in the hands of advanced statisticians, predictive modelers, data scientists and others who have "the ability to connect science, art, technology and business acumen," Kierzkowski said.

The potential of machine learning and predictive modeling has changed how we look at and analyze our data.

The human difference that your team brings to the table -- positioning machine learning output in a creative, transparent and actionable form -- is invaluable. At Dun & Bradstreet, we have found that we are able to extract the maximum value from machine learning procedures by applying that layer of human intelligence -- the expertise and experience of our team. We are exceptionally proud of our analytics talent and love that machine learning tools can complement our employees rather than replace them. This also ties into my next suggestion.

Know your biases -- and ones in algorithms

In addition to using your analytics team to creatively augment machine learning in business applications, it's crucial that you use your team to govern the ML process in order to avoid analytical bias. The biases in machine learning initiatives can be both human and algorithmic. As humans, we may find it challenging to recognize our own biases, but we are capable of doing so nonetheless. Machines can't recognize their biases -- yet.

Workforce diversity helps to minimize bias in machine learning algorithms, and it can't just be left to chance. As writer Alex Hickey noted in a March 2018 article on the CIO Dive website, "The onus of forming a diverse team to train these algorithms, which has been shown to reduce bias in these systems, falls on management."

One of the benefits of having a team comprised of people from different backgrounds is that they can cross-check the machine learning algorithms for latent biases that a more homogeneous team might not be aware of, and then teach the algorithms how to avoid these biases. This is especially true when it comes to making sure that the output from a machine learning initiative doesn't favor one region or group over another due to unseen biases in the original programming.

Establishing what you want to accomplish, leaning on your human assets and vigilantly watching for biases are three basic ways to optimize your integration of machine learning in business processes, but they're all worthwhile steps to take. For Dun & Bradstreet, the potential of machine learning and predictive modeling has changed how we look at and analyze our data, and I'm very excited to see how deep learning and other advances continue to change and shape our business.