Insights Matter, the Data Proves It
In the column "Big Data, Big Deal" (June 2012, CRM), I described what big data is all about and what to consider when it comes to its challenges. In an informal way, this is actually the second half of that article. In this half, I want to focus on how to get insights from big data since, when you boil it down, if you don't gain anything from big data, it's a big nothing.
Gaining insight from big data requires a lot of work and a lot of thinking prior to, during, and after the work. You'd be way off if you thought that the tools are going to give you the insights. What the tools do is give you the information you need to gain the insight. If we were dealing with a single customer, it might not be that difficult, but we are potentially dealing with insights about hundreds, even millions, of customers, since many of the customers are looking for personalized responses based on their individual expectations and behaviors.
I'm going to tell you a story that will make it all clear. Telenor is a Norwegian telecom with 203 million customers across Europe and Asia. It is the 17th largest mobile network operator in the world, with $17 billion in revenue in 2011.
Like any telecom, Telenor has churn problems. But the churn is a result as much from what's called promotional churn—low-cost offers from competitors—as it is from dissatisfaction with the services it provides. The difficulty with this kind of churn is figuring out what offers keep customers, which can be an expense unless you do some modeling with all the data you have on customers. A blanket campaign to effectively keep customers might be somewhat costly and generate churn itself.
Plus, the company was dealing with 203 million customers, of which 150 million were mobile, which is where most of the churn occurs.
So what could Telenor do? It had tons of data, an undifferentiated target group, and a definite problem to resolve.
Telenor did the smart thing. It applied what is called uplift modeling to the problem.
Unlike traditional modeling, which only focuses on a target group, uplift modeling focuses on both a target group and a control group. Its purpose is to measure against the control group to predict both the likelihood of a particular customer performing an action and the change in the likelihood of that customer performing the action—meaning it looks at context, time, and the impact of other customers or influences and influencers on the target group. So, for example, it would not only look at the percentages of the customers likely to churn, but also at the impact of a campaign on those customers' likelihood to churn.
Uplift modeling identifies four groups.
Sleeping Dogs: Those who would leave due to receiving an offer. They want to be left alone.
Lost Causes: Those who are leaving regardless of anything.
Sure Things: Those who are staying regardless of anything.
Persuadables: Those who might stay due to an offer being made.
What Telenor did was run analytics to create models that identified the individual customers in each group. The company used data from billing systems, customer service data, and sales/purchase data. The variables they looked at were plan size, number of products, and call volume, among many others. Telenor also looked at the propensity to leave and the responsiveness to offers of each customer.
Note that all the data I've spoken about is traditional structured data, not social data. IBM's Institute for Business Value, in its 2012 study "Analytics: The Real World Use of Big Data," found that the vast majority of businesses used traditional structured data for their modeling and other big data work—up to 88 percent of the respondents used transactions as their primary source. We're still at an early stage when it comes to how we handle big data.
Regardless of sources, the value to Telenor should be pretty obvious. The only group worth targeting is the Persuadables because, according to the model, they could be prevented from churning by a good enough offer from Telenor.
The results were superb. Not only were costs reduced by 40 percent due to the highly specific targets, but uplift modeling yielded less churn on these large quantities of data over traditional models. Traditional models reduced churn by about 5 percent, but uplift modeling enabled the company to keep an additional 1.8 percent of customers, a 36 percent improvement over traditional models. That might not sound like much, but remember the number of customers Telenor has. Churn reduction at 6.8 percent saves millions of dollars.
Clearly, there are benefits to big data. But to garner the value, you have to take some steps to make sure that you do it right.
Here's a simplified version of the process that can lead to insights from big data.
- Don't treat it as big data. Identify the types of data and the data sets that you are interested in. You don't need to scan all the crowdsourced data in the world. You want something from it—remember, you are looking for an outcome, not just reviewing data. Perhaps you are trying to resolve an issue…in the case of Telenor, how to reduce churn.
- Decide what it is you're looking for. Perhaps you are hoping to find out something about your brand; even more germane, to find opportunities for your company, even when your brand is not being discussed—something often overlooked. In the case of Telenor, the company was trying to find out who to focus on to reduce churn.
- Extract specific information for insights. Use all the data sources you deem necessary, be it traditional structured data or unstructured social data or even video data if it makes sense. Telenor used the traditional structured data en masse to find its target audience for churn reduction.
- Run the analytics. Focus on predictive analytics, not just business intelligence–descriptive. Telenor did uplift modeling, which is predictive.
- Develop insights. In Telenor's case, the company learned to target the Persuadables.
- Take action. Telenor gave the appropriate offer and reduced churn by 6.8 percent.
Think about this. We have the data available to us and the sources to choose from and the tools to help us get insights from the data. We are in a world now that provides no excuse to not pay attention to the needs of your customers. When you have thousands or even millions of customers, you can make a difference by acting on what you learn—and that's predictable.
Paul Greenberg (@greenbe on Twitter) is president of consultancy The 56 Group (the56group.typepad.com) and cofounder of training company BPT Partners. He is also the conference chair of CRM Evolution (www.destinationcrm.com/conferences/2013/). The fourth edition of his book, CRM at the Speed of Light, is available in bookstores and online.
Driving Lifetime Relationships
Customers are not like cars, but your business should be
Big Data, Big Deal
What you don't know can hurt your business.
How Will CRM Evolve?
Changing the game with predictive analytics.
Beyond Purchase Histories
Predictive analytics drive high-value lifetime customers.
Birst Bolsters Predictive Analytics for Sales
Predictive analysis of transactional data improves sales performance.
Changing the Face of Workforce Optimization
Predictive and real-time analytics offer tremendous potential for contact centers.
Use the Scientific Method to Tackle Big Data
A step-by-step approach to marketing success.
SMBs Seek Enterprise-Class Predictive Analytics
Simple and adaptive solutions can reduce costs and improve profitability for small companies.