Using a machine learning solution allows the ERP to imitate the manager’s decision criteria, and then allows the managers to focus on approving those decisions and managing the exceptions.
The Imitation Game was applied to computer science by Alan Turing and became known as the Turing Test. (If you are not familiar with Alan Turing, check out the book “Alan Turing: The Enigma” or the movie “The Imitation Game.”) The original objective of the game was for a machine to successfully imitate a human being under a written interrogation. This was not to prove that a machine could think, just that the machine could be taught to imitate human conversation. Of course, the machines took a huge step forward in understanding and imitating human language when IBM’s Watson defeated the best human experts in Jeopardy. But imagine the Imitation Game applied to ERP: could the ERP make business decisions that effectively imitate management decisions, leaving humans to approve the decisions and handle the exceptions? If the management decisions made by the ERP – how to allocate scarce inventory, how to schedule jobs in the plant, which orders should ship first – were indistinguishable from the decisions made by management, could we say that the ERP had successfully imitated management?
Sound farfetched? I’m not so sure. I’ve been bringing my analytical skills current with online classes from Coursera (if you’re a programming geek like me, check out this Machine Learning class), and can report that the combination of algorithm, data, and computing power available today has the potential to completely transform what we think of as ERP. The field of Machine Learning offers solutions that could learn from past management behavior and accurately predict future decisions. And many of these approaches will start appearing in ERP solutions in the near future.
For example, a manufacturer that cannot keep up with demand faces a problem of allocating scarce inventory among hundreds of product-hungry retailers. Every day, someone must decide who gets product and who doesn’t. Conventional solutions attempt to implement naïve rules that must be known in advance (e.g. all ‘high priority’ customers get inventory first), but inevitably managers spend hours first gaming the system (after all, which customers should be ‘high priority’?), then more hours each week manipulating the simplistic recommendations to try to achieve an appropriate balance. These decisions just aren’t as clear cut as we would like.
We face decisions that are not so clear cut in our everyday lives, like choosing what songs you want to play from your favorite music streaming service. Your service pays attention to your decisions about what to listen to, assembles a playlist of recommended songs, and allows you to select or ignore each one (using those decisions to refine the next playlist it will present). You don’t have to select anything in advance, the machine learning algorithms that assemble the playlist can work solely based on your past decisions. Any other information you can add (your age, gender, favorite band, etc.) only improves the results.
But back to our management dilemma: who gets the scarce inventory? A machine learning solution focuses on capturing the decisions made by management to predict future decisions – much the same way the music service focuses on capturing decisions about which songs you play in order to predict what you want to hear. Using a machine learning solution allows the ERP to imitate the manager’s decision criteria, and then allows the managers to focus on approving those decisions and managing the exceptions.
The Machine Learning field has developed technology that can leverage enormous quantities of data about decisions to truly imitate the decisions made by the experts – without the experts ever defining even the most basic decision tree (let alone one that would yield all of the same decisions). Machine Learning systems can be “trained” with historical decision data, develop decision models that accurately reflect the criteria used by the experts (even when they don’t know their own criteria), and accurately predict the decisions those experts will use next week when they allocate scarce inventory again. Some algorithms hold the promise of learning on-the-fly the changing preferences of the experts, allowing the allocation recommendations to change with the business climate.
If you’re still skeptical, remember that we use systems driven by machine learning all around us. From the traffic routing systems in your GPS to facial recognition systems, from Siri to the suggestions presented in the Google search box, every day you are interacting with systems leveraging machine learning. Expect your ERP to start leveraging Machine Learning techniques in the near future, and see if you can tell whether the decisions recorded in your ERP were made by management or the system.
Written by Lane Nelson, “…for the first time in close to a century it’s probably easier to make a fortune in auto manufacturing working with (or founding) a startup than by signing-on with one of the big companies.”
Written By Henry Nelson. The customer focus matches HarrisData's culture and promise throughout our 45 years as an independent software vendor. Until a customer realizes benefits from our software, we have done nothing.
Written by Lane Nelson Using a machine learning solution allows the ERP to imitate the manager’s decision criteria, and then allows the managers to focus on approving those decisions and managing the exceptions.
Written by Lane Nelson The introduction of cloud-based (and cloud-priced) solutions have muddied the accounting waters again, such that a $4M software implementation project must be expensed if the customer is implementing a cloud application, but can be capitalized if implementing the same software on premise.
Written by Lane Nelson For some, an organization running ‘hands free’ enterprise applications is as hard to imagine today as it would have been to imagine an assembly line with no workers in Henry Ford’s day.