You are here

You are here

4 tips for adopting enterprise machine learning

Ben Lorica Chief Data Scientist, O'Reilly Media Inc.

Interest in machine learning has grown steadily over the years, and many organizations are aware of the potential impact machine learning tools and technologies can have on their business.

But the reality is we are still in the early phases of adoption, and the majority of companies have yet to deploy machine learning across their operations. In fact, since the introduction of machine learning models at scale during the dot-com boom, it's taken nearly two decades for ML models to become mainstream.

To understand more about how machine learning has progressed, O’Reilly recently issued the results of a new survey that explores the state of machine learning adoption in the enterprise. The findings suggest that only 15% of the 11,000 respondents work for companies that have extensive experience using ML in production.

So, for the large majority of companies just starting their machine learning journey, what are the first steps? One is to explore the ways that sophisticated users—in this case, those who work for organizations that have been using machine learning models for more than five years—approach this technology.

Here are some of the practices that can help organizations go from exploration and evaluation to implementation, so they can start deriving value from all machine learning has to offer.

Introduce more specialized machine learning roles

New roles—such as data ops specialists and machine learning engineers, who specialize in building and deploying machine learning models—are starting to stick. In fact, nearly 40% of sophisticated organizations have a designated ML engineer, double the number of organizations just getting started.

Not surprisingly, organizations with more experience deploying machine learning models to production are more likely to use these newer job titles, which also include data scientist, data engineer, and deep-learning engineer.

If history is any indication, the emergence of machine learning-specific roles is indicative of the hype around the technology. Just take the role of data scientist, for example: The title was coined around 2008 to describe practitioners who worked on data projects, and now more than half of all companies surveyed have hired for this role. Organizations that are hiring for machine learning roles now and putting resources behind its deployment will be better poised for the future of business.

Implement specific machine learning success metrics

Less experienced organizations rely more on product managers or executives to determine the criteria for machine learning project success, but sophisticated organizations entrust their data science leads to set team priorities. Additionally, respondents who belong to the most advanced organizations are likely to use multiple success metrics, which may include business metrics, ML metrics, and statistical metrics, and they measure for bias and fairness.

For example, over half (54%) of respondents who belong to sophisticated companies check for fairness and bias, compared to less than a third (32%) of companies that are just starting out.

This is another indication of how machine learning adoption introduces challenges that diverge from the standard practices of software engineering—and thus there's a need for leaders who can navigate the waters. This type of expertise helps organizations set parameters and build better models that generate better business results.

Approach machine learning model building differently

As machine learning becomes more widely used, many organizations are adapting the processes they've used in software development to build data products, such as agile methodology and Kanban. However, some experts have pointed out failures in those approaches and offer advice about how to work around them.

A key mindset shift required to address these issues involves understanding that machine learning model development is different from software development. As such, completion of the machine learning model-building process doesn't automatically translate to a working system.

The data community is still building tools to help manage the entire lifecycle, which also includes model deployment, monitoring, and operations. While tools and best practices are just beginning to emerge and be shared, it's still early days for model lifecycle management.

Build robust model-building checklists

Organizations that have extensive experience in deploying machine learning models have much more robust model-building checklists than their peers, already including checks for transparency and data privacy. In fact, 53% of respondents who belong to companies with extensive experience in ML check for privacy. This is especially important now that most companies need to comply with European Union's General Data Protection Regulation (GDPR).

GDPR mandates "privacy-by-design," or the inclusion of data protection from when a system is first designed rather than as an addition after the fact. This means more companies must add privacy to their machine learning checklist. Fortunately, for most organizations getting started, new regulations coincide with the rise of tools and methods for privacy-preserving analytics and ML.

Get going now

Machine learning has enormous potential, but in order to reap the benefits, it's important to put your organization in a position to take advantage of all of it. By hiring for machine learning-specific roles and leadership, implementing success metrics, and building robust model-building checklists, your organization can start to advance on its machine learning journey.

Keep learning

Read more articles about: Enterprise ITIT Ops