If an organization is only interested in finding out which customers are about to close their accounts, then it may not matter what data their AI uses to make decisions. The business can just let the AI take over, let it respond to at-risk accounts based on the data science, offering discounts, free services, or even a teddy bear — whatever it takes to save the account.
Increasingly, however, organizations don’t want to be in a constant state of reaction. In other words, these organizations are interested in the “why” that their data can offer. The math behind a decision based on an individual’s address, eye color and height may be correct, but it is using data that the organization can do nothing about. Until we understand “why,” it’s not time to turn decisions over to an AI; we need to apply machine learning and data science to get more insights.
Here are some very common variables used to analyze churn:
- What steps preceded the point that put the customer at risk of churning?
- How often are they interacting with the company?
- How are they interacting with the company?
- When was the last time they were on the website? And for how long were they active?
- When was the last time they used a company product such as a bank’s credit card?
But these variables also share a common problem: businesses struggle to get value from them. Organizations need to be enabled to make preemptive changes that alter paths before accounts ever approach closing. This requires using machine learning provided insights that have business value.
The challenge faced by machine learning is that at its most basic, it is a mathematical analysis. It does not recognize that events can vary in value as differentiated by whether or not the organization can interact with that value. What’s needed are models that reflect not just quantity, but quality of interactions.
Such input only comes from people. Specifically, it comes from business people who have the experience, practical knowledge and intuition to recognize that customer height, for all its apparent predictive power, is not something that the organization can affect, while things like frustrating bill payment and login web pages can be improved.
This is not to suggest that data scientists who develop models that include height data are not themselves human! They almost certainly are. It is simply that their training and focus is on predictive power, not necessarily the reality of actionable business insights.
That is why in order to get the most out of machine learning, data science and business experts need to work together, putting the business in a position to change the way the data science team operates. If your business wants to apply data science successfully, then the business needs to play its part in the process.
That requires a change of perception and attitude. There is a false notion that machine learning projects are all about specialized data skills, SQL queries and Python code that data scientists excel at and business users tend to shy away from. In reality, both are needed. Both parties need to overcome their discomfort, sit side by side, and contribute to the process, putting the data scientist at the keyboard working with the data, while the business person looks over their shoulder and says, “I can’t do anything with that variable,” and “Why not include this other variable?”
Three Areas Where Businesses Can Inform Data Science
In what areas can the business contribute to data science?
- Data selection: The business can play a role in selecting the data sources that it can actually do something about.
- Feature engineering: The business can help close the gap between the mathematical creativity employed by data scientists in building variables to describe data and practical variables that have meaning to the business.
- Interpretation: Models will always throw combinations of variables that are inscrutable to the data scientist. Business users can spot the relationship between, for example, how long it takes to pay a bill and what type of credit card the customer uses. In this case, the business user knows what the data scientist does not — that different credit card types see different bill paying pages.
For most business applications, the hunt for the golden unicorns who understand the math, the data, and the business has put an artificial limit on the potential return of machine learning investment. Breaking down the barriers between data science and business and putting multiple minds with varied expertise to work on models is more affordable and more productive.
If a business wants to get the most out of machine learning, it’s time to stop treating data science and machine learning as special projects that exist in a bubble. Instead the business needs to step forward and take an active, ongoing participatory role, feeding machine learning with daily doses of real-world business insight.