Google has been using machine learning to improve its business analytics for years, and judging from all the recent excitement about the technology, you’d assume enterprises everywhere would be following suit. Yet, the reality is somewhat different. According to some studies, only 22 percent of companies are already using machine learning’s analytical power by implementing algorithms in their data management platforms.
So why aren’t more organizations taking advantage of the science that can make them “Google smart”? Probably because they believe it’s too complicated. They might think their data is too voluminous or unreliable, might consider the required data preparation too time- or resource- intensive or might assume only data scientists have the skill set needed to leverage ML. While those concerns may have held true a few years ago, today, all those objections can be overcome.
The Renaissance of ML
In the past, using ML algorithms was complex, and the outcomes could be perplexing and unpredictable -it was difficult to understand how the technology classified data, so you were never sure what type of results you might get. However, the rapid adoptions of the cloud and new technology tools have combined to help simplify ML and make it more accessible to a broader base of IT professionals. Let’s take a quick look at how some of these tools help.
Machine learning is data driven, which means you need a lot of data to make it work. And, to make it actionable –for example, to deploy it for uses such as personalizing real-time customer experiences — you also need to process the data in real time. Furthermore, ML requires a LOT of computational power, particularly to learn models. Fortunately, the cloud is ideally suited to deliver on these requirements and can also play an integral role in simplifying the use of ML and making the technology more affordable and manageable. Additionally, there are also now a variety of commercially available tools that lower the barrier of entry and the complexity of ML while still working natively with the languages and frameworks used by the technology in the cloud. For example, recently introduced drag and drop components give line-of-business (LOB) developers and power users the tools to complete many of the tasks needed to leverage machine learning’s power without the complex coding. Some examples include:
- Providing pre-packaged math and statistics ‘routines’, e.g. boiler plate code which previously could only have been created by statisticians,
- Assistance in preparing or “featuring” data (for instance, many statistical models only work with numeric values, so these tools help by converting strings into numbers with numbering schemes or assigned ranges, etc.), and
- Help training and validating models (a process that previously required deep knowledge in methodology and math).
ML’s Many Advantages
Of course, while ML has considerable raw power that can now be used more easily for data insights, we are not yet at a point where we can simply plug in data and get instant results. Raw data still needs cleansing to weed out flaws and exceptions, because until trained, ML can’t recognize them. While machine learning technology is still being trained, the most effective way to employ it is to combine it with the human expertise that can recognize data issues and exceptions that machines can’t. This approach ensures that organizations take full advantage of machine learnings’ unique prowess: the ability to become smarter over time.
In fact, to gain a true competitive edge with ML, it’s essential to put some control into the hands of business users – the people who really know the data and can bring their human insights to bear. By combining business users’ insights with ML, companies can quickly leap forward in their analytics. Once they begin the process of making their ML smarter, they’ll start to experience exponential gains. Conversely, companies that aren’t employing ML will find themselves miles behind those that have invested the human equity.
Once the complexity associated with using ML has been stripped away, and companies have deployed resources to train the technology, a much broader base of business and IT professionals can leverage it to analyze larger data volumes and more data types than ever before. Businesses can now more easily employ ML for operational insights, to uncover patterns humans can’t find and to look at countless data variables. Massive volumes of data can now be more easily deployed to explore operational analytics such as:
- Which products are likely to be bought together? (Collaborative filtering)
- Will an event happen in the future? (Classification)
- How much, what will be the number of…? (Regression)
- Who are my gold customers? (Clustering)
- What will be the price of this stock in a month? (Gradient boosted tree)
- Is fraud occurring? (Decision tree)
- Is that image a known intruder? SVM (supervised learning)
Of course, it goes without saying that ML is only as good as the data you feed it, so those using it need to create a solid data governance plan. ML can do a great job of automating good decisions, but it can also dramatically increase the negative impact of bad decisions made using inaccurate data. In fact, data can even be used to intentionally infer alternative facts. Establishing guard rails against these potentially negative impacts is a must, particularly in light of growing data privacy concerns in the U.S. and abroad. For example, General Data Protection Regulation (GDPR), an impending EU regulation intended to strengthen and unify data protection for individuals, establishes the right of explanation. Under GDPR a European citizen has the right to ask for an explanation regarding the output from an algorithm that’s been employed to make an automated decision that impacts them, such as being refused a loan or being disqualified as a job candidature.
It’s not as far away as you think
Thanks to advances in technology, infrastructure frameworks, and growing cloud adoption, companies no longer need to rely exclusively on data scientists to reap the benefits of ML, they just need the right tools – and then, they too, can become “Google smart”.