Can Customer Retention be Automated by Machine Learning?

Machine learning is a big topic and one that has people excited. One would be hard-pressed to find an analyst group that doesn’t anticipate enormous value from machine learning in the coming years. But when it comes to practical applications, it’s open to considerable interpretation, or more precisely, misinterpretation. If one wants to evaluate just what machine learning offers for customer retention or any other business operation, we need to clarify what it is and what it does.

Let’s start by borrowing from Gartner’s IT glossary which says, “Advanced machine learning algorithms are composed of many technologies (such as deep learning, neural networks and natural-language processing), used in unsupervised and supervised learning, that operate guided by lessons from existing information.”

That last bit, “guided by lessons from existing information,” is the key. Machine learning is not science fiction. Machine learning is evaluative. It can identify who is at risk of churning. A sufficiently sophisticated machine learning system can also tell why someone is at risk, especially if that system has access to data about the customer journey.

Machine learning predicts who will churn

But will machine learning tell you what to do about that risk? Yes…and no. It can tell you whether a customer who meets a certain, very detailed profile is likely to respond to a $15 coupon if the system knows that such a program has worked before. Machine learning will not, however, create a recommendation out of whole cloth. It can’t tell you whether a new strategy such as offering that same customer a $100 gift card or even a cruise would retain them (if those options were financially feasible) if there’s no prior observation to support that conclusion. Keep that in mind. That fact is significant to contextualizing machine learning’s role.

There’s also a tendency to think that the latest advances in machine learning--currently neural nets and deep learning--can finally solve the entire churn problem, uncovering both who is at risk and what to do about it. But while models are getting better and better at predicting who will churn, they can’t come up with creative solutions.

Machine learning’s ability to tell us who will churn alone is very valuable, creating a strong ROI for using the technology. It will help identify at-risk customers better than any individual can. It can quickly find novel connections between data points. Human-powered analytics might eventually deliver the same answers, but will require more effort, more time, and as a result will be accomplished less frequently. Those are very high value reasons to incorporate machine learning into your retention efforts.

So why not let machine learning run the show?

The intent of machine learning is to generate rules. Data goes in. Machine learning then uses that input to create rules in complex ways. Criteria for success and failure are defined. People might never notice a logarithmic relationship between the duration of a phone call and a customer’s happiness. Machine learning software tirelessly and without bias tests all those variables and relationships, which is how it can deliver novel results. The rules applied by machine learning often don’t make sense to human minds, but the fact that they are counter-intuitive doesn’t make them any less accurate. But, it is still up to people to determine their practicality and decide how--and whether--to apply those results or simply react to the model’s final score for a particular customer.

Ultimately, machine learning will make recommendations about what to do based upon available data within strict criteria. Juggle the dice, that is, change the order of the entries in the data set or otherwise alter parameters, and the recommendations will be different. Having human checks in place brings necessary knowledge, insights, and creativity into play.

Why you need machine learning

Machine learning is not a panacea, but it should be in place now. Machine learning is iterative by nature and that characteristic points to its best application to customer retention. The longer you use machine learning, the more lessons learned and knowledge fed into the system, the greater the return will be. The immediate return may be small, but the benefit is cumulative and takes time to mature.

Machine learning is a powerful tool for business analysts. The system can tell you that customer A has an 80% chance of closing an account and customer B a 30% chance. It can deal with a shifting, changing, and ever-growing array of data to deliver that analysis. And it can do this with surprising accuracy, far more reliable than a gut feeling.

And why you need human creativity and input

If you’ve pinpointed with great accuracy who will churn, now what? What do you do in response to that scenario? Machine learning does not answer that question. You can create a system with boundaries and a canned set of responses. Only you can decide if your business can afford to offer discounts of 10%, 25% or more vs. the cost of an extra agent call or a simple email. More importantly, how many of each of these offer types can you afford and which groups of at-risk customers should get which discount? These criteria, like fashion, inevitably change with time and in unknowable ways. If you remove the human element entirely, you lose the dynamism and creativity that is the hallmark of long-term progress.

In the end, whether machine learning can or will automate customer retention is the wrong question. The question is how best to implement machine learning now within the context of multifaceted customer journey data in order to derive the greatest return.

Robert Bagley is the Chief Research Officer at ClickFox.


Don't forget to check out the blog about the impact machine learning can have on the customer journey!

Machine Learning Part 1

 

Written by Robert Bagley

Robert Bagley is the Chief Research Officer. He oversees the innovative application of Data Science practices for client engagements and product features, charting future analytic and technology strategies. Prior to his current role, he held various leadership positions at ClickFox focused on building and managing top-notch delivery teams in engineering services and customer support. Before ClickFox, he developed applications focused on improving key processes and reducing operating costs at the Georgia Institute of Technology and Southern Company. He holds a Bachelor of Science in Electrical Engineering from the Georgia Institute of Technology.

Subscribe to the ClickFox Blog

  • Be the first to know about what's trending in Journey Science
  • Stay up to date with our latest Journey Analytics Products
  • Be a part of the Journey Science movement!