Machine Learning in the Enterprise


Machine learning in the enterprise is already happening. Sophisticated predictive sales applications can pinpoint the best prospects to close, with much greater accuracy than experienced sales reps. Recommendation engines may be suggesting additional products that customers might want to buy. In banking it is already commonplace to have systems which process loan applications, and even decide on the most appropriate interest rates. Deposit rates can also benefit from intelligent agents that process large amounts of historical data to find a rate that is acceptable and profitable on a case by case basis.

Business managers will never need to understand the internals of machine learning, but it is certainly worth knowing what it is capable of, and how it is finding its way into many business applications. The essence of machine learning is that a suitable algorithm will consume data from the environment and learn how to optimize a given outcome – usually profitability, but also revenue, customer satisfaction, or a combination of these. So in the case of loan approvals the machine learning algorithm will process large amounts of historical data to establish which conditions lead to the most profitable loans with least risk. Variables such as age, other loans, salary, mortgage payments and so on will figure strongly. This cannot be a one-off process. The environment changes, and as it does so the algorithm needs to learn how these changes are affecting performance against the objective that has been set.

There are many types of machine learning, and some flavors are more appropriate than others in particular industries. The highly regulated banking and insurance sectors must provide transparency, and this does limit the type of machine learning that can be used. Decision trees and tables are commonly used, since the decision logic can be easily interrogated by regulators. More complex and sophisticated methods are used in predictive sales applications, where ensemble methods create hundreds or thousands of solutions which can then be ‘averaged’ to create a consensus. The accuracy of these methods is often surprisingly high. Machine learning is also starting to make its way into applications that take input from sensors. Applications already exist for the detection of faulty items on a production line. Image recognition is one of the fastest areas of growth in machine learning, and is finding widespread use in biometrics, manufacturing and security.

Machine learning is the science behind other high profile activities such as data mining and predictive analytics, and is one of the essential techniques used in artificial intelligence. The qualifying property of machine learning is that a system learns as it gains experience from increasing and more timely data. If it doesn’t, it isn’t machine learning but possibly some form of static statistical analysis or static rules specifications. The advantages of systems that learn as opposed to static specification should be obvious. After the 2008 financial collapse a whole new class of people experiencing financial stress was created. These were people who would previously have been given high credit ratings with the associated privileges. A static rule based system would have gone badly wrong for applications such as loan approval. In a machine learning environment, the system would have quickly learned that profiles had changed, and more appropriate ratings given.

Having said that machine learning is an essential component of artificial intelligence, the scope of machine learning in business will expand very quickly. Artificial intelligence is concerned with the creation of intelligent agents – an agent being something that performs some sort of useful work. An agent is considered intelligent if it performs well against some pre-defined objective. Artificial intelligence has already made its way into law practice with an artificial intelligence lawyer that will execute much of the legal research. This is not a future but is already used in several large law firms.

For large organizations with data science teams, machine learning will provide opportunities to gain real competitive advantage. Highly bespoke intelligent agents will find unique ways of dealing with customers, suppliers, trading partners, and even employees. Platforms for the teams who carry out this work are increasingly found in the cloud, avoiding set up costs and the staff required for support. It also means that sophisticated models which require large amounts of compute power can dynamically access resources as they are needed. Machine Learning as a Service is the name given to these platforms, and some also act as market places where businesses can sell their own solutions and purchase those of others.

In reality however, for many businesses machine learning will be largely invisible – buried in applications that automate various business processes. Obviously a business needs to be able to manage the various agents created through machine learning. To this end it is necessary that the agents are managed within a structured environment. For many machine learning applications, we find that the resulting agents are automating, partially or completely, the decisions within the business. This is certainly the case with loan approvals, detection of faulty items on a production line, and product recommendations. To this end it is necessary that businesses adopt a Decision Management practice, where agents are documented, changes audited, performance monitored, and other functions that ensure the decisions made by these agents are well understood and optimized. The alternative could be truly alarming.

Machine learning, particularly within the context of artificial intelligence, will affect most functions within many businesses – large and small. Even human resources is expecting a significant impact from machine learning. And from supply chain management, to fraud detection and reducing customer churn, machine learning is already outperforming human decision making. This is happening very rapidly. We are not projecting five or ten years out here. It’s happening now, and some businesses are already lapping those that have yet to wake up.