BETA
This is a BETA experience. You may opt-out by clicking here

More From Forbes

Edit Story

3 Steps To Jumpstart A Machine Learning Strategy

Following
This article is more than 7 years old.

In every area of technology, people are discussing machine learning and deep learning, especially for the Internet of Thing (IoT) market. This week’s announcements from Google and Salesforce also focused heavily how machine learning and artificial intelligence will be used to improve how we live, work and play. It’s clear that machine learning will become part of new products and leading SaaS offerings.

As companies look to build data-driven business processes, this functionality is critical. The question is how will companies unlock data and insights from existing products and how should business leaders approach this new concept of machine learning? To answer these questions, I spoke with Sara Gardner, CTO for Hitachi Insight Group. Gardner said there are three core principles that business leaders should follow to get started in the world of machine learning.

Similar to any IT or business project, IT and OT should start with defining a significant business problem that analyzing large data sets can answer. While this sounds obvious, many companies struggle with selecting a business goal that matters and defining the scope of the goal to make it achievable. For example, we need to reducing costs isn’t a clear enough direction. You have to frame the problem into something data can solve. For example, a mining company that may want to decrease the out of service days and the repair costs associated with truck equipment failure. Given that a pair of truck tires for a mining vehicle can cost hundreds of thousands of dollars, a good place to start would be discovering the issues that lead to tire failure. Machine learning for predictive maintenance saves money and minimizes outages.

The next step is defining what data are available to your organization. Many companies already collect vast amounts of information but haven’t tapped into those resources yet. Others will have to instrument equipment with sensors to collect data. Machine learning requires learning from experience, and in the case of certain equipment, you may have limited failure data to analyze. In this case, you need to create failures using simulators. Given the amount of data an organization could have access to, Gardner noted it’s important to identify the key indicators surrounding the problem you’re trying to solve. For example, changes in tire pressure would be a key indicator in the mining use case – Machine Learning techniques help here too.

As part of step two, you also need to seek out tools that can ingest massive sums of data and help you visualize important trends within this data. Gardner noted this is one of the reasons that Hitachi acquired Pentaho, a company that provides big data tools to extract, prepare and blend your structured, unstructured and semi-unstructured data. It also provides visualizations and analytics.

Given Lopez Research’s discussions with enterprise customers, I recognize that there are many types of tools a company could use to do this. The challenge is selecting the right tool for the right job. And vendors like Hitachi also realize the importance of having this category of products in its toolkit.

Now we’re entering the toughest part, step three. Your company understands the goal, and you have the data, but someone must define a set of methods to analyze that data. Your organization must address issues such as how to come up with a standard schema. The technology team must define what type of algorithms make the most sense for the kind of problem you’re trying to solve and type of data that you have. Understanding these methods is where the ever elusive data scientist comes into play.

What I’ve learned from discussions with many vendors and end user IT shops is data science is part art and part science. Exploration and trial and error are part of the process of creating algorithms. Machine learning implies that you are building systems that are continually learning and adapting the algorithms based on what it has learned. This type of analytics differs wildly from the batch data processing and weekly business intelligence dashboards that many companies use today.

But what if you don’t have a stable of data scientists at the ready? A company must look to its suppliers to fill the data scientist gap. I’ve noticed that more companies are trying to address the lack of data scientist by providing a set of out of the box algorithms or solutions for their customers. In the case of Hitachi, it is working on blueprints and solution cores for its primary industries. These offerings can range from machine learning algorithms to hardware for processing data. The solution core should provide upwards of 60 to 80 percent of the required functionality.

A company can either augment these solutions with its staff or work with a services company to complete the solutions. Others such as Salesforce.com are embedding the processes of data analytics into apps such as its CRM and service cloud. Meanwhile, Oracle has a series of IoT PaaS services that integrates with its existing business applications. Since there’s no one size fits all to data analysis, companies should use a combination of these solutions to meet the analytics needs of various parts of the business.

The final piece of the strategy is defining what do you do with the insight. Domain expertise is just important as the math. Without the domain expertise, it’s hard to apply the data trends in a way that maximizes business value. For example, there’s a cost to taking a piece of equipment out of service for maintenance. There are usually several options of how you can approach this problem. You could reduce the time the equipment operates or increase the frequency of maintaining the equipment to lessen downtime. If you have an early warning system, you may extend maintenance windows.

The conclusion: Machine learning tools will provide the potential to unlock new insights and drive business value. However, machine learning isn’t a solution that you can simply purchase and just drop into your organization for insight. It requires experimentation and alignment between business and IT to drive efficiencies and competitive advantage.

Follow me on Twitter or LinkedInCheck out my website