Machine learning is a big buzzword right now. If you’d like to learn more about what it is, we have a blog series you can read about machine learning techniques.
But if you’re interested in using your big data for machine learning and starting a real project, this is the blog post for you.
We talked to Brian MacDonald, Data Scientist on Oracle’s Information Management Platform Team about tips for success for your machine learning project.
Our tips for machine learning success include:
Let’s get into it.
We most often see success with machine learning projects when the company has a very specific business problem and they’re willing to do anything that’s needed to solve that problem.
Brian explained, “Now one of the companies I worked with wanted to create a better churn model because they were losing customers rapidly, which was affecting profits. They had a lot of data, and they believed they could use machine learning to help them identify the customers who were about to leave. If you have a $100 million problem, spending $30 million isn’t a big deal.”
That’s the kind of company that tends to see success. Having that laser focus and need to find a solution helps ensure success because there's no other option.
On the other hand, some organizations have problems in general but they’re not sure how to go about it. They might hear the buzz about machine learning and decide they can use it. But although machine learning can do wonders, it can also be very complex and is rarely an easy fix. Once they discover that, these organizations sometimes decide against going forward. Or if they start, they sometimes founder midway.
When you’re going into your project, you need to keep the end goal in mind at all times. Take a look at our big data solutions if you want inspiration.
This is related to the first tip above. But the number one factor we see for success is having high-level execs who support your machine-learning initiative. You want a C-suite that says machine learning analytics and data are important to your business. If you have that support and vision, your program is much more likely to be successful.
On the other side, when your program is driven by IT, what you tend to hear is something like, “We’ve never done this internally and I don’t know how to sell it.” This type of approach is less likely to be successful, not because the technology isn’t working, but because of internal issues.
For example, because machine learning is in essence automated decision making, sometimes people can view it as a means to replacing their own jobs. If employees at your company are worried about replacing jobs or lowering the head count they, you’re not going to get a reception that’s as strong. And keeping this in mind is important, because then you can decide how to counter this kind of attitude.
But that’s another reason why having executive support is so important. It becomes a way to go around that attitude, more easily.
In essence, you often need backing and money to make a machine learning initiative a true success.
When you’re starting out, you’ll want to start with a concrete business benefit like increasing sales. That’s an example of a business benefit that’s tangible, that everyone can see, and which won’t take too long to identify. The length of time it will take really depends on your goal, but it should be less than a year. Some say it should take no longer than four to eight weeks for the project to prove success. If there's no success, then it's time to move on.
If you don’t have a business value that’s measurable, question why you’re doing this because at some point you’re going to have to justify your project.
Some people might say things like, “We think machine learning is the future” or “We need to develop those skills.” Well, that’s investing in building skills and R&D for the future, and that’s a business benefit. However, whether you have the assets to spend on that kind of research really depends on your company size and corporate strategy, and you should really try to align with that before you start.
Here at Oracle, we’ve been fortunate enough to see many success stories with machine learning. But here’s one example from the energy industry that stands out.
This company is a leading supplier of systems for power generation and transmission, and is one of the world's largest producers of energy-efficient, resource-saving technologies.
This company purchased Oracle Advanced Analytics, which is also available in the cloud and part of the Autonomous Data Warehouse, to help them add predictive modeling capabilities to the services they offer to their customers.
They were successful in large part because they were so focused.
They had a very specific business problem, they got their executives behind the goal, and they identified short-term, measurable business benefits. There’s another item you might want to add to that list: purchasing the right machine learning technology, which can often contribute greatly to the success of your project.
So choose carefully and wisely, and contact us if you’re interested in our machine learning capabilities. We're here to help you make your machine learning project successful.
And if you'd like to try building a data lake and use machine learning on the data, Oracle offers a free trial. Register today to see what you can do.