Remember that AI or machine learning is not the right solution for every type of problem.
There are certain problems that machine learning solves really well.
But there are also other problems that machine learning really isn’t the best, or even the right, solution.
When not to use machine learning?
We’ll give you 1 very easy way to know whether you should not use machine learning.
If you can describe exactly the rules to solve something, don’t use machine learning.
A clear signal not to use machine learning.
The key here is that if you are facing a problem that can be solved, 100%, with pre-determined steps that you can articulate, then you should solve it using traditional automation – by programming those steps into a system or app.
For instance, you don’t need machine learning to:
- Calculate and tabulate your accounts
- Automatically respond to email enquiries with a template
- Allow your customers to update their billing information
- Schedule an appointment with your client
- Take an online order placed by your customer
When to use machine learning?
So when is it a good time to use machine learning?
There are 2 very strong signals that AI or machine learning will help you. If you see either of these, you may want to consider machine learning.
1. You cannot describe the rules
If you cannot describe exactly how you would solve something, but you actually can solve it pretty well, then machine learning is a very good solution.
The 1st signal for when machine learning is a great solution.
Many tasks that need a human to think about, are often very difficult to describe as rules.
For instance, it’s difficult to know:
- Will that sales enquiry turn into a sale?
- Is my client likely to continue keeping me on retainer?
- Is that email a phishing email?
The commonality here is that you cannot describe the rules to give an answer.
Therefore, these challenges cannot be adequately solved using a simple, rule-based solution.
And a lot of factors and variables could influence the answer. In fact, you may have difficulty even listing any but the most obvious factors.
Or, perhaps you could think of some rules, but realize that many of these rules overlap and interact in ways that need to be adjusted very carefully.
When you see this pattern, you can use machine learning to effectively solve this problem.
It is a little counter-intuitive. But it’s actually easy to recognize.
2. You cannot scale
If a problem needs a human to solve, but there’s a lot of them to solve, then machine learning is a very good solution.
The 2nd signal for when machine learning is a great solution.
For instance, you may be able to decide if that specific sales enquiry is likely to turn into an actual sale, but relying on your instinct and experience. And you may be able to spot that a specific email is an obvious attempt at trying to phish your password.
However, when you try to apply this to hundred or thousands of sales enquiries or millions of emails, it quickly becomes tedious.
You cannot maintain your accuracy, and you cannot process more than a small amount each day.
You cannot scale.
And… one more advantage
Machine learning solutions, like all forms of automation, are incredibly good at handling large-scale problems.
However, machine learning has one more advantage – it learns.
The more data it sees, the better it gets (assuming the learning algorithms are set up correctly).
This means that the larger the scale of your problem, the more accurate machine learning becomes, the better the answers you get.
