Click to learn more about author Ben Lorica.
Machine Learning (ML) use in production dates back about 20 years. Leo Breiman chronicled a sea change in Data Analytics in the notable 2001 paper, “The Two Cultures,” based on the introduction of Machine Learning models at scale during the dot-com boom. Even so, it took two decades before Machine Learning moved beyond the specialized “unicorns” of the internet to become a mainstream practice.
Today, Machine Learning has become more widely adopted in the enterprise, but how are businesses approaching this work? Do companies with more experience deploying Machine Learning in production use methods that differ significantly from organizations that are just beginning? For companies that haven’t begun this journey, are there any best practices that might help?
A recent survey, “The State of Machine Learning Adoption in the Enterprise,” seeks to answer those questions. The findings indicate that organizations well-versed in Machine Learning demonstrate a sharper focus on the road ahead by implementing key senior leadership roles, thinking more about the issues of privacy and bias and are better at measuring success than those businesses just getting started.
The emergence of Machine Learning-specific roles is one indication that Machine Learning is taking hold. For organizations with experience deploying Machine Learning, there is a notable impact across senior leadership teams that reflects this activity. Of these organizations, 81 percent employ a Data Scientist, 20 percent employ a Deep Learning Engineer and 39 percent have at least one Machine Learning Engineer – almost double the amount of those organizations just getting started.
More advanced organizations are also savvier when it comes to ensuring model fairness and bias. Overall, 54 percent of respondents indicated that their organizations check for fairness and bias, compared to 40 percent overall. This trend carries over to privacy, with a 10 percent increase between companies with Machine Learning experience that regularly check for privacy (53 percent), versus those with less experience that do (43 percent).
For organizations in the European Union (EU), General Data Privacy Regulations (GDPR) mandates “privacy-by-design,” which states that organizations must ensure the inclusion of data protection from the onset of the designing of systems, rather than as an addition. This means more companies in the EU will add privacy to their Machine Learning checklist and, as Machine Learning models become more widely deployed and used, interest in transparency, interpretability and explainability will only continue to grow.
Machine Learning sophistication also impacts how Machine Learning models are built. One in two (51 percent) respondents reported that they use internal data science teams to build their Machine Learning models, whereas use of AutoML services from Cloud providers is in low single digits. This split grows even more pronounced among sophisticated teams, who tend to have Data Science leads set team priorities and determine key metrics for project success that would typically be performed by product managers in more traditional software engineering settings.
When it comes to measuring success, once again, respondents who belong to the most sophisticated organizations are likely to use multiple metrics to measure success, from improved decision making and engagement, to operational efficiency and revenue. Because these organizations are more likely to build Machine Learning models in-house and have more experience deploying models to production, Machine Learning is leveraged more carefully and only when there are clear benefits for doing so.
This research shines the light on some of the key learnings from deploying Machine Learning, and also where other companies should focus as they begin their journey of Machine Learning adoption. More advanced organizations have the important role of paving the way for others to deploy Machine Learning in the most efficient, productive and innovative ways possible.