How to Use Machine Learning in Business? 3 Steps

Rating

How to Use Machine Learning in Business? 3 Steps

When you're trying to analyze big sets of data, you might wonder how the new analytical and automation tools such as Big Data, Artificial Intelligence, and Machine Learning might help. Which kind of problems can be solved by machine learning? To decide that you should think about the problem at hand and the feasibility of the available data and your expectations from that data regarding the problem.

WHAT IS MACHINE LEARNING?
Machine learning is a computer learning process in which a set of various statistical methods along with programming are used to find patterns of predictability in a data set. Machine learning is a major application of artificial intelligence (which is a somewhat broader concept) that uses data and provides the ability to learn and improve from the previous set of data. Machine learning is great when it comes to finding how certain aspects of the data are related to the problem, but what it cannot do is access any information outside of the data which you are providing.

THE PROCESS TO USE MACHINE LEARNING. STEPS
Here are the 3 steps to use machine learning effectively to tackle a business problem:

1. ASSESS WHETHER YOUR PROBLEM REQUIRES MACHINE LEARNING
   Using machine learning can automate your processes but it's not always necessary as not every automated process needs machine learning.
   When the problem at hand is really simple then automation is feasible without machine learning. A simple problem means the task where there are predefined steps that are currently done by a human and can be replaced by a machine. This type of automation has been happening for a long time.
   But suppose if a problem is complex where it involves encoding the human language into a structured data set (suppose you want a machine to improve and not to make the same mistake again) then it won't be solved by the simple set of rules. These kinds of problems (complex ones) require learning from data and here begins machine learning.
   So to consider which problems should be solved by machine learning, those problems should meet the following two criteria:
   
   1. The problems need prediction rather than normal inference.
   2. The problems are well self-contained or do not depend much on the outside environment.
2. ASK QUESTIONS AND LOOK FOR MISTAKES
   Now that you have classified the problem as a machine learning problem and have collected the data related to it, check your intuition. Now even if it looks like it does miracles, machine learning is also just a statistical tool and instead of trusting it blindly, ask lots of questions. You will have to get comfortable with how the methods work. Once you ask enough questions and get enough answers and know how this will work out you will notice that it's not that magical.
   Just as every human makes mistakes that goes for machine learning algorithms too. There will be times when there will be mistakes because problems are complex. Some errors are the ones that you might not have anticipated. The more complex the error you find is, the better the algorithm will get.
3. DECIDE HOW TO MOVE FORWARD
   Now the last step is to decide what level of errors are acceptable in your problem. Is it OK to solve your problem with 90% accuracy and a 10% error margin? Or do you want a margin of error of only 5% or even 1%. Is there a certain kind of mistakes that may never be made?! You should be clear about these needs and expectations because that's what the algorithm of your (internal or third party) programmer is going to provide you. Make sure you agree on these things before you start.

Source: Anastassia Fedyk (2018), "HBR Guide to Data Analytics Basics for Managers", pp. 111-120

		Comments by date▼ Comments by rating Rating
		1 Mikey Wood Turkey
	Automated Data Science But this is where automated data science closes the gap with traditional machine learning. With automated data science, and data discovery, you can access external data for your machine learning projects, models and use cases. That is the power of automated data science and what it means for machine learning and for machine learning use cases as well.

		1 Jaap de Jonge Editor, Netherlands
	Risks of Using Machine Learning in Business Products and Services Business offerings, services and products that employ machine learning (computer programs that absorb new information and then change how they make decisions) are proliferating. Think of investment software, software analyzing whether a patient possibly has cancer, and self-steering cars. However, even the most advanced systems don't aways make the right, accurate or ethical decisions. There are 3 main causes for that risk: 1. Their decisions are based on probabilities. 2. Their environments may change in unanticipated ways by the algorithms that were initially established by the programmers. 3. Their complexity makes it difficult to determine whether and why they made a mistake. To avoid these risks, managers should: 1. Decide whether to let a system continuously evolve or use locked versions at time intervals. 2. Test the system thoroughly before and after it is rolled out. 3. Monitor the system continuously once it's in operation. Source: Boris Babic, I. Glenn Cohen, Theodorus Evgeniou and Sara Gerke, "When Machine Learning Goes of the Rails", HBR Jan-Feb 2021, pp. 77-84