Machine learning is one of the most exciting new technologies. Having applied this technology to some of our own systems, we’ve learned a thing or two along the way. So, when people ask me how they could use it to help their business, I turn around and ask them five questions to help determine whether they should use machine learning:
What is the problem you are trying to solve and does solving it have high value to your business or customers?
It probably should go without saying, but surprisingly, people often start thinking of “what” they will implement without really vetting out “why” they are trying to do it in the first place.
Do you need machine learning to solve the problem?
Machine learning takes a huge investment of time and data and offers no guarantees that it will solve your problem. Many real business problems today can be solved with Robotic Process Automation or other widely available technologies.
Do you have, or can you gather, the data that may reveal the insights you need to solve the problem?
Machine learning depends on creating a statistical database model that gives a prediction. To do that, you need a big set of data – in fact, the more the better. However, not all data is created the same. The most likely data sets to succeed in ML are those that are consistently formatted, deep in volume but limited in scope, and typically residing in a modern database technology.
Do you understand the data and can you use it to create a deterministic data model?
Subject Matter Experts are essential here because knowledge of the data is key. A data scientist can help you create a model, but they can’t divine meaning from the data you are looking at.
At FIS, we considered whether machine learning could improve reconciliations and how trades match up with other trades. We needed to look at how users were determining a trade break: the steps they went through, the information they were looking at, the attributes on the trade field they were assessing in order to decide. Importantly, we did not take all the data – only the data points we believed to be relevant to what we were trying to do. The result was a predictive model that could suggest the next step to a user and with time and more data, automate that step entirely.
Can you take an action based on the results of your ML model?
It’s great if it tells you something with accuracy, but that “something” must be information you can action; it has to provide with you some value, such as a way to automate a decision or change an investment strategy.
But if all those things line up, then there’s an excellent chance that you can leverage machine learning to its full benefit and solve whatever problem you’re trying to address.
Tags: Technology, Innovation, Investments