The craze for machine learning (ML) is growing rapidly. And although many organizations already understand how important AI will become in their future success, they still struggle to identify the right opportunities.
That, and understanding what problems AI and machine learning can successfully solve for them.
So then, what actually constitutes a good machine learning use case?
These are the Four Key Steps to Identify A Good ML Use Case
To identify a good use case for ML, you need to consider the following four questions:
- What do you want to predict?
- Is there accessible data that exists and that you can work from?
- What changes will be made by having a prediction?
- What impact will this actually end up having?
#1 Identify What You Want to Predict
You first have to understand the problem that you are trying to solve.
By knowing the problem that you actually intend on solving, it becomes easier to narrow your focus on finding the ideal solutions for the right problems within your organization.
Let’s look at the example of an airline.
The problem that they’re facing is that customers aren’t buying as many tickets online.
So the goal is to predict or determine any possible solutions that can offer new ways to either motivate customers to make more online purchases or drive more sales in other areas.
By asking the question “Do we know what might be preventing them from making online purchases?” helps us know what we’re aiming to predict.
The airline wants to be able to successfully predict and identify which of its customers have issues with online purchasing and what might be preventing them from doing so.
By answering this, we now know what we want to predict and we can better execute a plan of action.
#2 Take What You Can From Your Existing Data
To predict anything, the data needs to include measurements of what it is that we’re looking for. Otherwise, we end up wasting time on predictions based on test or mock data that we’ve never observed. Which doesn’t solve any real-life problems.
Using the airline example, the historical data set that we then use to train an ML model would probably list customers who have purchased online and those who haven’t, along with their purchase habits.
The goal is to know historically, which customers have made previous online purchases and if there are any correlations that we can apply to those that haven’t.
Once the AI understands the type of customers that make online purchases, it then has a data set it can create in order to work from and train its ML model.
#3 Identify the Changes that Will Come From the Predictions
We need to ask ourselves what changes might come into effect by making a prediction.
Simply having a prediction is not all that important. Predictions are like ammunition, without making proper use of them, they act as ornaments.
Because the prediction in and of itself is not that valuable, the action that gets taken based on a prediction is what makes a difference. The main purpose is to introduce change to an existing process.
In the case of the airline, a non-online customer might be assigned different offers, sales funnels to convert them, surveys to identify their barriers and any other forms of support that will help increase the chances of them making an online purchase.
By understanding what changes you’re able to make within the organization based on the prediction, the airline now knows what actions to take based on their prediction.
#4 Identify the Impact of The Changes On Your Business
Finally, we want to understand what sort of impact these changes have and the effect that the prediction will end up having on our organization.
The most obvious would be around driving additional revenue or decreasing costs. You could have an increase in things such as productivity and brand awareness.
For the airline, they want to make sure that they can measure any bit of impact that they’re going after. So this would mean looking at key metrics, such as increased sales rates or attracting more clients.
So for the airline, measuring these increases in customer retention or purchase rates will show them what impact this change is going to have.
Identifying key opportunities to implement AI is easy if you have the right questions.
So by using a checklist like this, each organization now has a better chance of being successful with AI.
Already identified a few Machine Learning use cases? Need to talk to the right people?