How To Minimize Customer Churn with Predictive Analytics
It is no secret that customer loyalty is immersive for any business, as acquiring a new customer is seven times more expensive than retaining an existing one. That being said, customer loyalty is mostly about choosing the right customers. After all, there is no organization that has never lost a single customer but there are many ways to at least minimize customer churn. To that end, following the release of a white paper titled How To Address Churn with Predictive Analytics, our media reporter Laura Myers chatted with Kenneth Sanford from Dataiku as he shed light on how organizations can employ predictive analytics to drive targeted marketing campaigns and product enhancements to minimize customer churn they will inevitably face:
First and foremost, Kenneth divides churn into two basic types, subscription and non-subscription churn. To explain subscription churn, he gives an example: “If you're a cable company or a newspaper you might have subscribers that are always coming in or leaving your portfolio of customers and since they pay every month, you know when they come and when they go.” Non-subscription churn, on the other hand, happens when users or customers can end their relationship with your business at any time. The tricky part about the latter type of customer is to predict churn, as their purchase frequency may gradually fall down in time, or they may all of a sudden never buy again.
According to Dataiku, the first step of combating churn is building predictive models that accurately identify and predict churn. From there, organizations should develop targeted marketing campaigns around preventing it or make product changes. Therefore, we asked Kenneth about what organizations would miss out on if they didn’t put a good churn model in place to address it. “One of the big problems of not having a good churn model is that you don't know whether a customer has actually left your portfolio in the case of non-subscription churn, or it's just a natural amount of time between purchase events,” he explained.
From a marketing perspective, a study shows that mid-to-large B2B organizations lose an estimated $958 million each year because their content marketing has gone wrong. This figure came as no surprise to me knowing 65% of content is unused mostly because the content is not findable or relevant. To reduce churn and minimize the waste, marketers need to assess their marketing programs in light of data, to make sure that the investments are delivering value to the business. To make that assessment right, they need a good predictive model in place which is easier said than done. To pick his brain on that matter, we inquired with Kenneth about the common pitfalls or missteps that can limit the effectiveness of a churn modeling project. He put into these words: “There are lots of places that a churn model procedure can go wrong. Obviously, you have to get the data all in the same place, so there's typically a lot of siloing processes in a company. However, in the modeling process, you have to understand that analytics is a constantly evolving organism, and because of that, you often have to have a lot of trial and error. For instance, a predictionj model you have created might turn out not to be very successful, then you need to go back in the loop again to understand why.”
It is important not to be discouraged at that point as Dataiku also indicated in the white paper, getting data into a usable format is a step of the process can account for up to 80% of the total time spent on the project. Or on a deeper level, as William Blake puts in, “The true method of knowledge is an experiment.”
Predictive analytics have also become the backbone of marketing automation platforms to translate hundreds of millions of data points into meaningful and actionable insights in order to track consumer behaviors from the perspective of what leads to buying decisions, so the marketer can find the most influential way of conversation and execute churn prevention campaigns. With the technology we have today, it is not hard to tell for marketers what sequence of events such as receiving an email, reading an article, or watching a video leads customers to buy or not to buy the product/service. However, to employ an effective modeling that can yield this sort of insight requires a good deal of patience as Kenneth pointed out during the interview.
You Have A Good Churn Prediction Model – Now What?
As a business, let’s imagine, you have a good churn prediction model in place which enables you to say, identify everyone who has higher than X likelihood of churning. Then, what’s next? If your company has limited resources where should you allocate them? Does it make sense to put time and energy into a customer base you're more likely to lose or, should you keep your focus on the customers that have already shown a more heightened engagement with your brand? Going big doesn’t make a lot of sense for two primary reasons. First, if you haven’t optimized the churn prediction model yet, attempting a “Big Bang” may set you up to fail as a lot of things can go wrong and your team could easily be overwhelmed. Secondly, as Kenneth stressed during the interview “analytics is a constantly evolving organism” so allocating your resources on a relatively smaller portion may save time and money. At the end of the day, a staggering 80% of a company’s future revenue will come from just 20% of its existing customers. With this in mind, prioritizing the most profitable or predictable portion may be enough to make a huge difference in terms of customer acquisition.
Naturally, one would like to target the people who are most likely to churn but this is not always the best idea. There is another parameter that marketers certainly need to incorporate; who's most likely to be convinced not to churn. For instance, it is a very common practice with fine brands to provide a promotional coupon along with a refund when a customer returns the product due to a defect because the data-driven marketer knows what approach is the most effective, he/she can customize, say, an additional promotional coupon that is very relevant to the customer’s purchase history to turn that dissatisfaction from the defected product into a successful customer service story. However, in this example, the likelihood of churning is not that subtle. There are many cases when you lose a whopping amount of customers before your eyes and that’s where a good churn prediction model should come into play.
Now let’s imagine that you have determined which of the likely churner to target, then what’s your game plan? Is it offering special discounts? If so, how will you cut through the noise and stand out from the literally hundreds of product/service providers to make your offer visible? Deciding on the promoted product and timing is also a huge factor to motivate a predicted churner to make a comeback. The key to open that door is, once again, data. As we all leave a trail of data whenever we interact with sites, finding answers to these questions shouldn’t be as hard as it used to be for today’s data-driven marketers after all. In the advent of technology, marketers now, for instance, can turn off ads to a specific customer when they have a current service issue as there is nothing more impulsive than seeing a promotion about the company when customers are experiencing an issue with them. Or, since marketers know what customers already purchased particular products, now those customers can be retargeted with a promotion for complementary products.
Deciding which of the likely churners to target and how to reach likely churners is not always an easy task. Therefore, one of the suggestions Dataiku makes in its white paper is uplift modeling, which categorizes customers into the four quadrants illustrated below in an effort to maximize impacts and minimize wasted money and effort:
Today, with the ongoing explosion in the volume of data collected, many organizations are living in irony when it comes to translating all the data into meaningful actions. The irony is gathering too much data but knowing too little about what this data represents. To address this emerging issue, many analytics platforms powered by cognitive computing technologies have started going beyond conventional text analysis. Dataiku, for instance, offers Dataiku Data Science Studio (DSS) to enable the beginners with machine learning algorithms to run open source algorithms to predict churn in a clickable interface without having to write any code. With DDS, data scientists are able to simply upload their data, select parameters, and have the platform automatically train and evaluate different machine learning models. More and more large enterprises including GE, L'Oreal, AXA, and HostelWorld have been employing this "collaborative" platform. According to Florian Douetteau, CEO at Dataiku, the data science software provider has doubled its revenue every year since its founding back in 2014. In fact, on Wednesday, Dataiku announced raising $28 million in a Series B round led by Battery Ventures, with participation from FirstMark, Serena Capital, and Alven.
Sending tone-deaf social media posts all over the place is no longer a marketing activity. At least, not a meaningful one. To develop a continuous churn prevention strategy, organizations should be able to gain insight at a fraction of the cost and time without the restrictions of traditional methods and take the guesswork out of business processes. While developing an effective churn prediction model, it is mission critical to bring cross-functional teams together. Once you have found who can be saved from the churning out, special deals may stop the bleeding wound just like a band aid in the short run but in many cases, sooner or later, the wound reopens if you do not address the root cause. There is not a one-size-fits-all approach as every product line has its own unique audience and message but there are always so many ways of building trust with your customers in an authentic way without losing revenue. All you need is a great team armed with predictive insights to uncover hidden opportunities as well as competitive threats.