By Rich Clayton, Vice President, Business Analytics Product Group
You’ve probably been hearing a lot of marketing buzz recently about “machine learning.” More and more companies are using this new set of technologies that combine data mining and predictive analytics to automatically discover patterns in the data and to learn from those discoveries.
For example, the first time you shop on a retail ecommerce site, you purchase a home brewing kit. The algorithms behind the scenes learn that you enjoy making your own beer. But they also have a ton of other customer data to pull from—in some cases, millions of customer profiles—many of whom also enjoy home brewing.
Based on the likes and preferences of those other home brewers, the site can offer you products that home brewers tend to buy (for example, beer mugs, salty snacks, or even a local football jersey for the big game).
McKinsey defines machine learning as “based on algorithms that can learn from data without relying on rules-based programming.” In the past, artificial intelligence was mostly rules-based; business people would develop the rules, so that if the algorithms encountered a certain condition, a specific response would automatically be triggered.
A simple example: your credit card company might detect a transaction under the amount of $5. Then, within 10 minutes, it detects another transaction for $200. The credit card computers are programmed to identify that pattern as fraud. That is a business rule that a person programmed into the application.
Machine learning would take this a step further, asking, “Were these two transactions processed in separate locations? Did they both take place in stores where the card holder has never shopped before?” Machine learning compares the pattern to other, past instances of fraud to identify the likelihood that the credit card was hacked or stolen.
Machine learning has been around for about 50 years. Probably not a day goes by where our lives aren’t impacted by it—digitally and physically.
Today, however, machine learning is rapidly approaching a tipping point. This is because of the reams of data that companies and vendors can harness to compare patterns and learn. (Oracle, for example, has so much anonymized data that we can offer data-as-a-service (DaaS) to our customers—more on that shortly.)
There are 5 factors currently influencing the rise of the machines:
The business opportunities have only just begun to scratch the surface of what’s possible. But with more machine learning, companies also face more risk. A simple example of unintended consequences is price discrimination. A machine cannot make moral judgments about discrimination; it can only make decisions about classes of customers—with no understanding of who is part of a marginalized group or what the legal implications might be.
Because of this risk, CFOs must pay more attention to machine learning. As more decisions are automated, the risk of running afoul of laws and regulations goes up. One poor decision by a machine might not amount to much, but in the aggregate, they can have a material financial impact.
In the end, it’s the CFO who will have to answer for any fiscal impact. Finance leaders carry the fiduciary responsibility of the company; it’s important for them to understand how these algorithms are operating, what decisions they’re making, and what the ramifications might be for shareholders.
We recommend that finance officers undertake four steps to harness machine learning and control the associated risks:
Machine learning really is about experimentation, hypothesis testing, and automation. A data lab allows companies to test out their machine learning algorithms and assess the potential risks, before they are put into production.
A data lab is not the same as the data warehouses of the past. (Some CIOs call it a “data lake.”) Rather, it is a platform for innovation, where developers can bring together a discrete data set that hasn’t been tested before and use machine learning to identify hidden patterns.
For example, if my company manufactures windows with sensors embedded in them, can I take the anonymized data from those sensors and sell them to security companies, police, or firefighting organizations? How can I predict when a window or door is going to fail, and then sell that data to maintenance and repair companies?
This kind of specific insight would only come from a data lab where people are chartered with innovation and learning. Once identified, these patterns and predictions can be commercialized to add value to the business.
The National Health Service in the United Kingdom delivers healthcare to all 65 million citizens of the UK. Last year, they set up a data lab to learn more from its huge volume of patient data. Within 3 months, they reworked the health card application process, using anomaly detection to find fraudulent activity.
By showing value in a relatively short time, the NHS received backing to expand. They now have a long term strategic goal of saving £1 billion over 5 years.
A data ethics committee can audit algorithms for unintended consequences, thus reducing the risks associated with machine learning. Algorithms need auditors, just as much as (if not more than) any other area of the business.
According to Gartner, “By 2018, 50 percent of business ethics violations will occur through improper use of big data analytics…. Failure to properly understand and mitigate the risks can have a number of unintended and highly impactful consequences. Those can include loss of reputation, limitations in business operations, losing out to competitors, inefficient or wasted use of resources, and even legal sanctions.”
Before you put any of your data lab findings into practice, examine all the potential pitfalls—including any legal, financial, and brand implications.
There are vast amounts of data out there. Today, even small to midsize companies collect far more information than they can ever commercialize—and as more and more devices are connected to the Internet of Things, the amount of data transmitted will grow exponentially.
By far, most of this data will be just noise. Only a fraction of it will ever be predictive. So, decide which data might be worth something, and which data sets you can discard. Not all data needs to be stored. Some of it you might need to keep for regulatory purposes; others, for commercially useful predictions and products. Keep what you need and what is potentially valuable. Otherwise, you will be overwhelmed with data that adds no value.
The World Economic Forum predicts that, by 2020, 5 million jobs will be lost to artificial intelligence. The question is, "What new industries and jobs will be created?"
For finance professionals, it means automating as much as they can today, so that they can move away from the traditional accounting tasks of performing transactions, reconciling accounts, and compiling reports. With these tasks automated, CFOs and their teams can focus on partnering with the business to analyze available data, identify new business opportunities, and provide strategic guidance.
Oracle’s work with professional accounting bodies such as CIMA and AICPA has shown that finance professionals are already moving toward this new role, identifying new key performance indicators for changing business models and making recommendations on where to invest next. CFOs should double down on these analytical skills, since machines are likely to take over most of the routine tasks performed by accounting departments today.
Drawing on Oracle DaaS, Oracle has embedded adaptive intelligence into its portfolio of cloud applications. These AI apps will provide recommendations for smart offers to your customers, the best candidates to fill job openings, the best freight value for shipping, or the best payments terms to your suppliers. These recommendations are presented in-context while your employees are using Oracle Cloud Applications.
The algorithms continuously learn as you use them, to provide the best recommendations and practices to optimize user experiences to help your business evolve.
For finance professionals in particular, the Adaptive Intelligent Applications will amplify Oracle ERP Cloud to help them negotiate the best supplier terms and optimize cash flow—especially during high-volume periods such as end-of-quarter.
As machine learning continues to evolve, finance leaders should make the effort to familiarize themselves as much as possible with the business opportunities it creates. The winners will be neither the machines alone, nor humans alone, but the two working together effectively to create new products, services, and value as yet unimagined.