Identifying Business Processes That Can Be Machine Learning-Enabled

What to look for and expect when analyzing workflows for tasks can be automated with Machine Learning

Welcome to another installment in my series on 6 steps to apply machine learning to your business! Assuming you’re all caught up on the difference between AI and ML, we’re ready to move on to step two: identifying business processes that can be ML-enabled.

In this article, I will help you do exactly that by building an understanding of:

The relationship between automation and Machine Learning;
The benefits and challenges you can expect from ML integration;
Business process mapping and general workflow analysis.

First, let me share a story.

Back when I first joined my company, I noticed we had two people spending three full days to collect product data for the previous week. By the time the product owner reviewed the performance report, it was out-of-date and lacking accuracy due to human error.

Naturally, I thought to myself, Here is something we can automate! So I asked the technical teams about it and they firmly replied: “No.” They did give an explanation: “Our manual process is well-optimized. If there’s a solution we need, it’s adding more people.” (I wonder if you’ve heard this before?)

Being the over-eager newbie, I wasn’t about to give up. So I broke down the workflow and identified specific processes that could be fully automated.

Today, not a single employee works on data collection or report generation. I’m about to explain my process so you can learn how to make this happen in your company.

The Relationship Between Machine Learning And Automation

Before we can use the robot, we must understand it. In this case, the “robot” I’m talking about is just a mental picture. Imagine Automation is its body and Machine Learning its brain.

When applied to a workflow, Automation covers the majority but Machine Learning is the core. It is the “Artificially Intelligent” agent.

Pure automation cannot predict and analyze; we need ML for that. Keep this distinction in mind as we proceed.

How to Know if ML is Right for Your Business

“If you do not change, you can become extinct!”
Spencer Johnson, Who Moved My Cheese?

Change is the law of life. In the corporate world, stagnation and lack of innovation can run you out of business. As technology progresses, customers have come to expect better products. To stay competitive, you must embrace the advances.

Here are a few benefits of Machine Learning automation:

Efficiency. Scale up your business while removing tedious processes and minimizing without additional human resources.
Accuracy. Reduce human error and careless mistakes.
Speed. Data processing will be more accurate and faster.
Prediction. For user behavior and preferences.

But just because ML automation is all the rave and comes with some great benefits doesn’t mean you should start blindly integrating it into your business.

Some tasks can benefit from ML and others simply aren’t suited to it. So let’s talk about how to find out which is which.

Business Process Mapping

To begin, you must conduct a thorough review of your business to identify specific tasks and processes that can be automated with Machine Learning.

The best way to do this is by clearly define all of the processes your business undertakes through business process mapping.

Look at each process and outline:

How it is carried out;
Who is responsible;
To what standard the process should be completed;
How the success of the process can be determined.

Workflow diagrams can also help you outline the tasks involved in a particular process. Here is an example for Customer Support:

Customer Support Workflow Diagram, Source: Lucidchart

To further break down a specific task, we can often use this general workflow:

Granted, this might seem like a lot of work, but defining all the tasks in a process — and listing the workflow for each one — is exactly how you’ll uncover ML opportunities.

Thus, I highly recommend breaking this down in detail. Once you’ve done so, you can look at each element and ask: Can this be Machine-Learning automated?

Tasks That Can Benefit From ML

Notice the word “manually” in every stage of the general workflow above. This is a simple key to identifying processes that are:

Human-intensive: require a lot of people and time.
Highly repetitive: require the same task to be performed over and over.
Tedious: require manual processing of large amounts of data.

Looking back at our workflow analysis, we can see how all four processes can involve a lot of human resources, repetitive action, and tedious work — exactly what we’re looking for! Workflows like this can be automated and ML-enabled.

Let’s relate this to my story from earlier. Here’s a breakdown of what took the two employees three days:

Collect data manually: From various platforms and channels. (Automation)
Analyze data manually: Look for relevant information. (Machine Learning)
Make decisions manually: Interpret findings and choose what to present. (Machine Learning)
Take action manually: Create a report for the product owner to review. (Automation)

By breaking their workflow down into these stages, I identified that (1) Collect Data and (4) Take Action” could be fully automated and (2) Analyze Data and (3) Make Decisions could be Machine Learning-enabled.

Here are some more guidelines to help you determine whether or not a task is suited for ML:

ML may work well if the task:

Needs natural language interaction (e.g. Siri)
Calls for a personalized experience (e.g. item recommendation)
Analyzes huge amounts of data from different sources (e.g. weather forecast

ML may NOT work well if the task:

Involves creative action (e.g. drawing)
Requires 100% accuracy; has no allowance for error (e.g. a calculator)
Simply doesn’t want to be automated (i.e. you don’t want it to be)

Now let’s wrap this up!

The ABCs of Identifying ML Opportunities

Analysis: Which Of My Processes Can Be Machine Learning-Enabled?

The business process mapping, workflow analysis, and examples discussed above will help you answer this question. But before taking action, you’ve got to consider a few other factors.

Backing: Can I Collect Relevant Data?

Data is key in Machine Learning. Without relevant data to back it up, an ML model can’t learn. Thus, a lack of access to clean, usable data can make a task that is otherwise well-suited to ML unfeasible.

Questions to ask:

Where is the data stream?
Where is the data stored? (Is it in a third-party API?)
How much data can be trained? (More on this in future posts!)
How can the data be cleaned and unified?

Challenges: What Integration Issues Should I Consider?

Suppose you’ve identified a business process that’s perfect for ML, and you have all the data needed for the job. Now you have to look at the “external” factors, such as budget, talent, and company structure.

Questions to ask:

Do I have the budget to implement a new system structure, considering the cost of specialized labor, hardware, and data storage?
How will I measure the return and value of the ML application?
Do my employees have the digital skills to work with the technology?
How will I compete with giant enterprises and startups to attract top AI talent?
How will ML integration change the structure of my company?

In Conclusion

That’s it! You now have a way to identify the processes in your business that can benefit from Machine Learning. Once you determine what is feasible, you’ll be ready to start actually integrating ML into your system.

And that’s exactly what I’m going to help you do next: apply Machine Learning to your business. (Coming soon!)