Automation. Ever since the dawn of the Industrial Revolution, people have been automating work to make it more efficient, drive down costs, and relieve employees from the drudgery of mundane tasks. And the automation of today’s business applications is already quite sophisticated.
Now we’re on the eve of another Industrial Revolution as machine learning takes automation to another level, allowing computers to make decisions on our behalf. But how exactly does it work?
Automation in conventional computer programming is rule-based: If x and y conditions are met, then z can occur. It is very rigid and can take into account only a defined number of parameters. Let’s take invoice receipt as an example. A major retailer will receive thousands of invoices a year and each one must be matched to a purchase order. If the right goods have been received in the right quantities at the agreed price, then automatic matching can take place and the invoice can be passed for payment.
But there are many reasons why an invoice might not match. Perhaps a supplier has invoiced more than one purchase order, or acceptable alternatives have been substituted for the goods originally ordered, or perhaps there have been breakages in transit. If any of these conditions are met the matching process must be passed off to a human, who will more than likely make a judgment based on previous experience.
New breed of software
Machine learning is a new breed of software that can learn without being explicitly programmed. Machine learning can access, analyze, and find patterns in huge quantities of historical data in a way that is beyond human capabilities. The patterns can then be refined using supervised learning until the software has “learned” to identify a defined set of parameters and trigger the appropriate actions
For example, based on historic data the software can learn how the previous set of invoices was matched manually, and based on the historical patterns, come up with a sophisticated multi-variate model to match invoices, greatly reducing the need for manual intervention. This results in reduction in shared services resources and better optimization of the finance functional area resources to more value-added tasks.
To make machine learning a reality in enterprise applications not only requires a new breed of software, it requires advanced processing power to process the vast amounts of training data needed to make the algorithms intelligent.
SAP recently partnered with Nvidia to leverage the advanced processing power needed for machine learning. If you were unable to attend SAPPHIRE NOW live, you can watch the replay: Speed Digital Transformation by Bringing Machine Learning to the Enterprise: Our Partnership with NVIDIA.
The combination of the new breed of software with affordable supercomputing power is game-changing. For the first time, intelligence can be built into enterprise applications, allowing knowledge work to be automated. In addition, completely new use cases that could not be contemplated before are possible!