Big Data and Analytics in a Customer-Focused Enterprise: Inside Scoop with Karl Rexer

CustomerThink Founder/CEO Bob Thompson interviews Karl Rexer about the role of big data and analytics in customer-centric success.

 

• Data mining vs. analytics
• Business value of big data and analytics
• Customer attrition analysis
• Surprises from data miner survey
• Does analytics drive business performance?

Interview recorded August 7, 2012. Transcript edited for clarity.

Bob Thompson:
Hello, this is Bob Thompson of CustomerThink. For this edition of Inside Scoop, my guest is Karl Rexer, founder of Rexer Analytics, based in Winchester, Massachusetts. Karl is a leader in the field of applied data mining and he has very broad experience in a number of areas including strategy and leadership, predictive modeling, statistics, data mining, direct marketing, CRM and market research.

Today I’m very excited to chat with Karl about some recent findings from Data Miner Survey. We’ll talk a little bit about big data and what that means, and how companies can get an advantage from using analytics. Karl, welcome to Inside Scoop. It’s great to have you on our program.

Karl Rexer:
It’s great to be here.

Bob Thompson:
Tell me a little bit about your business. I know you have a lot of experience in many different areas. Can you tell me a little bit more about what exactly it is that you do with your clients?

Karl Rexer:
Sure. And also I should say that it’s not just me. I have a small staff and they are outstanding, as well. We’ve been going since 2002, and we provide data mining, consulting and analytic CRM consulting to our clients. So, that’s everything from fraud detection to predictive modeling for marketing or customer attrition, forecasting, all sorts of different analytic consulting and solutions that we deliver to our clients.

Bob Thompson:
OK, so, you have people—including yourself, I presume—who can build these predictive models and help with the implementation of them using software tools out in the market, correct?

Karl Rexer:
That is correct. So, we build the analytic algorithms to help solve these problems for our clients. But we also really try to work very closely with our clients to make sure that the analytic or mathematical models that we build are very tightly coupled to the business challenges that they’re facing.

Data mining vs. analytics

Bob Thompson:
I want to start and just talk about some kind of basic terminology that perhaps some people are confused about. Let’s start with something very simple. What’s the difference between data mining and analytics?

Karl Rexer:
Now that’s a great question because there are a lot of different terms out there—so, everything from predictive analytics, data science, data mining, even statistical consulting. And really, for many years, we’ve been doing largely the same types of things, but the terminology has evolved over the years.

I think that many of these terms are really almost synonyms, really doing pretty much the same thing, but some of the different terms try to stress different aspects of what they’re doing. Predictive analytics as a term tries to stress the fact that many times, the types of analyses are trying to predict something about what someone is going to do or predict something about the future. Other times, if people are doing other types of analytics—maybe something like customer segmentation—there’s not always a predictive component, and they might not use that term.

Bob Thompson:
Right. So, they’re just trying to understand what the data is telling them, get some insight from it, but not necessarily trying to build a predictive model.

Karl Rexer:
Exactly. And so, really many of the techniques, the analytic techniques being used are often very similar, but when applied to different sorts of areas or applied with a different focus or intent, oftentimes people will use different terms.

Bob Thompson:
I’ve heard the term over the years, “data mining” and then “text mining.” Is it basically—text mining, data mining, analytics—all kind of the same thing?

Karl Rexer:
I would say that largely they are, although you mentioned the term, “text mining,” and that is an interesting one that is a bit different and also a bit newer, I would say, in that much of the older data mining was not making as much use of natural language or text data. It was utilizing more numeric data. And the inclusion of text is something that is newer and an exciting new area that more and more people are pushing into because there can be oftentimes very rich information there that can be found sometimes in databases, sometimes lots of places that’s now just totally textual and needs different types of analysis to bring that in to utilize with the other analysis.

Bob Thompson:
It sounds like you can use the term “analytics” in a lot of different ways, and analytics and mining can mean roughly the same thing.

Karl Rexer:
Yes.
?

Business value of big data and analytics

Bob Thompson:
Perhaps one of the reasons why this term “big data” has caught on because, as I’ve observed the market over the last 15 years or so, is because there always seems to be need to come up with an acronym or some term that will at least appear to simplify things. Maybe that’s the key phrase—”appear to simplify”—because big data is what everybody is talking about now.

But when I look at some of the commentary, it still seems to come down to not just the data but the analytical tools. Text mining, data mining, predictive analytics are all things you can focus on big data. So, let me throw that back to you. Do you have a point of view on what big data is, what it really is—not how others are defining it—if one of your clients asked you? And then, perhaps more importantly, what’s the big deal? Why should a business leader pay attention to it?

Karl Rexer:
Well, first of all, I’d say that I have not seen any one agreed upon definition of big data. And so, first of all, it is what the definition sounds like—it is a large dataset. Now, what some people might call large, other people might not call large, so that’s why there’s not full agreement about that.

For some of our clients, we certainly have analyzed over a million US tax returns or tens of millions of bank transactions or grocery store transactions. Now, to us those seem like big datasets, and so to us that seems in a way to be big data and big data analysis. But if you were Google or Facebook, Amazon or looking at web traffic, or if you’re in a scientific field looking at some astronomy data or some genome research, you might have data that’s much larger and different. Sometimes it’s wide, in terms of lots of columns, or very deep in terms of the number of rows, and so other people’s data might be far larger than the datasets that we’ve been using.

Bob Thompson:
Right, and I know one of the solutions I’ve heard about is Hadoop, which apparently is designed for these massively large datasets—larger than the traditional ones that you talked about, in terms of tax returns and so on.

Karl Rexer:
Exactly. On the business side, what distinguishes some of these truly big data situations is often really the velocity of the data—how quick is that data coming at you, and the immediacy of the data as well. So, if you’re talking about airline traffic data or social media or maybe a shopping recommendation website, there’s immediacy. It doesn’t matter as much to analyze air traffic data last year. You’d analyze that right now for decisions you need to make right now, and there’s people’s safety relying upon it. So, that’s a big data situation, but it’s a big data situation where the information is coming in very rapidly and you need analysis to be turned around very quickly.

Bob Thompson:
OK. So, there could be applications, big data applications to deal with this high velocity. That’s one of the three V’s I keep hearing about—volume, velocity and variety. And then, in terms of variety, isn’t social media a part of the mix, too? It’s all this stuff out on the social web that theoretically could be analyzed.

Karl Rexer:
Absolutely. That actually gets back to the other part of your question and say, “Why should a business leader care about big data and analysis of big data?” And I’d say the reason for that is the potential for competitive advantage for your business by doing so. There’s both the internal data—and that could be the millions of transactions that I was talking about before, where the competitive advantage might be to better know and understand your customer so that you can react faster, provide the things that your customer needs and keep them as a customer. Then there’s also the external data. That’s social media, Twitter, the other things. You might be able to mine that to get information about customer sentiment.

And I have a nice example of that, I thought. Last year, I was the moderator at Predictive Analytics’ World Conference, and one of the speakers there gave a great case study about a mobile phone company that was introducing a new handset. They monitored the customer sentiment, the positive and negative comments that they saw overall across a number of different social media sites about the phone handset as they went through that launch. They saw that there was some negative sentiment because some customers seem to be confused about some features on the phone, and they were giving negative comments about it for that reason. But the company was able to respond very quickly and post step-by-step guides on their website and respond right within those shopping websites and blog responses to be able to turn the customer sentiment around. And that was, I thought, a great example of a company paying attention to that, doing some analytics and taking action to make the situation better for them.

Customer attrition analysis

Bob Thompson:
Great example. All right, let’s talk about another example. You mentioned that analytics and data mining can be used internally as well. Can you give me an example of how a company could take this massive amount of transaction data that it’s been collecting and get some real business value out of that?

Karl Rexer:
Sure. One of the ones that we’ve been doing for years for our clients has been doing customer attrition analyses and predictive models around customer attrition, as well, and that’s really something that has provided a lot of value to many of our clients. So, it’s really able to provide both some strategic value, and then companies can take action on that, but also some really tactical value to these companies, as well. And when they take action on that, they also are achieving good results.

Let me back up and say what this really is. I should say most companies have customers, and so when you have customers, some customers are likely to stay with the company and some are likely to leave or “attrite,” as we sometimes call it. If we can identify are there certain groups of customers that have commonalities that are more likely to leave and others that are less likely to leave, companies can use that information to make decisions.

On the strategic level, for one business-to-business (B2B) company, we did some attrition analyses and discovered that their clients in certain business industries stuck around and were really sticky clients. They stayed with them and they had long lifetime value because they earned revenue on those customers for many years.

However, there were other customers they had in different business industries that turned very quickly. They’d spend just as much to acquire those customers, but their overall lifetime value is much lower because they only stuck around for a year or two and then moved on, did other things, and they did not retain those customers. That company used our attrition analysis to change their focus for customer acquisition, so now they changed their acquisition strategy to focus more on trying to sign up those customers that are going to have longer lifetime value, those industries that are more valuable to them.

Bob Thompson:
So, your analysis didn’t necessarily explain why they were leaving, but it said, “Look, if you’re in industries X, Y and Z, for whatever reason, they’re more likely to leave. Maybe you ought to not spend so much money trying to acquire them.”

Karl Rexer:
Exactly right. So, right, our analysis was off their internal data, and no, it did not focus on why—although, when we presented those findings to them, some of the senior executives of the company said, “Oh, OK, and yes, we really understood.” They understood why that was, but they had never seen it quantified before, and so had never been able to use it when designing their strategies. It was very valuable for them to see it quantified so they could then see where they’re going to make the cutoffs of which industries they were going to go after and which ones they were not. They could also follow up with some customer surveys to ask people when they leave, why they leave, and then sometimes adjust their service.

Bob Thompson:
What about if they just said, “Well, look, we don’t want those customers to leave. Maybe we can find some strategies so that if we’re seeing indications in this transaction data that they are likely to leave, we’d like to do something to try to retain them,” is that an option?

Karl Rexer:
That’s a great question because that actually goes to my second point there. For other types of analyses that we do in the customer attrition space, we’re actually assigning a customer risk, an attrition risk score to every customer in a client’s database. This is a score typically on a 1 to 100 scale where if the score is higher, they’re more likely to leave. If the score is lower, they’re less likely to leave. It’s very much like your FICO credit bureau score, which is an indication of how likely a person is to default on a loan, but in this case here, the score is indicating how likely that customer is to close their account and leave.

We’ve built several of those types of models for our clients in a number of different industries—banking, investments, B2B, service companies of several different types. One of our clients recently boosted their annual revenue by about 11 million dollars because they were able to take action and reduce their attrition. The way they did that is they designed different customer outreach programs, different touch strategies for customers who had different types of attrition scores.

They also took into account the customer profitability—and again, we’ve done this with a couple different companies. But an example in the banking industry is that if you have a valuable customer and they have a high attrition score, they get a call from the branch manager. If you have a person who has a medium profitability but a high attrition risk score, they may get a call from someone at the call center or from a teller or something like that when they’re not busy, not serving customers in the branch. If you have a high attrition risk score but reasonably low profitability, they might not reach out to you at all and they might just let you leave, and that might be OK.

Bob Thompson:
What kind of factors would enter into this? I know these algorithms are pretty complicated, but what would be things that would tend to be built into a model that would help indicate whether somebody’s likely to leave?

Karl Rexer:
Well, the types of things that we see across different industries are—different measures of customer engagement can often be critical in these models. Now, each industry might have different measures of that, but for a bank, it might be how many transactions a customer is having.

Bob Thompson:
Of the number of transactions, it might indicate there’s something wrong, perhaps, in the relationship.

Karl Rexer:
It’s not the absolute number of transactions, it’s the trend. If they go from having many transactions down to a few, there might be something going on there, or sometimes it can be other things, like they might miss a payment or get assigned some fees because of the way they’ve changed using their services at a bank or at other types of companies, as well—so, if someone is late. Also, sometimes just the level of engagement can be important. So, someone who has numerous services is a more engaged customer than customers who have just one or two services with a company. That might not be something that’s changing much over time, but it’s just a measure of how engaged that customer is with a company.

Bob Thompson:
Yeah, that’s a great example. Wells Fargo is my bank, and I’ve spoken with them in a non-customer role doing articles in the past. They’ve really wanted to get customers to open multiple accounts, do different things with them, and I think they’ve been pretty successful with that. If you looked at all your customers and you found that some just had maybe one or two accounts, and that could be a factor all by itself in considering whether they might be likely to leave because they have less to close down and to actually move to somewhere else.

Karl Rexer:
Exactly. But the real power of these analytic models are that they can be multivariate models, so it’s not relying on any one of these things. And that’s where the predictive analytics is really different than reporting or writing SQL queries to look at different hypotheses that you might have about your data because you might start with hundreds of possible variables that might be used in your model, but you now want to narrow that down to which of those maybe dozen or so variables, used in combination, can help to assign a risk score to a customer?

Bob Thompson:
Yeah, yeah, it makes sense. I think that’s one of the areas that some business managers are a little bit confused. “Analytics” is used a lot, and a lot of vendors say they have analytics solutions. But often they’re really talking about tools to help somebody explore data, not necessarily a tool to help build a model to predict what’s going to happen. Predictive analytics is fascinating because business managers want to take action sooner rather than later.

Surprises from data miner survey

I did a little write-up recently about your data miner survey (report available for free download). Was there anything that popped out of the results in this last survey that really surprised you?

Karl Rexer:
I’d say that we’ve been doing this survey since 2007. And one of the things that wasn’t so much surprising to me this year because we’ve already been seeing the trend, but has been surprising to me overall across the multiple years we’re doing the survey, is the surge in the use of open source analytic tools, especially the R package, and the fact that it’s a surge not just among academic responders to our survey but among corporate users, as well.

Bob Thompson:
I’m familiar with R, and I know that’s open source. What would be another example or two of open source tools?

Karl Rexer:
Weka is another one that immediately comes to mind, that’s W-E-K-A. So, that surge in the adoption of open source tools in the corporate environments is one thing that has been a little bit surprising to me, and we’ve seen that now for the past few years. And then each year, there’s also little things here and there that surprise us because we try to ask slightly different questions each year to get at slightly different topics, and then there’s a core set of questions that we, of course, ask each year, as well.

One of the things we asked just last year that ended up being a little bit surprising to me was we asked about the data visualization tools that people were using, and I thought that among this group of data miners, we’d have lots of people using real data mining tools for their data visualization. But what we saw, reported by the data miners, were that Microsoft Office was far and above the one that people were using most to visualize their data. So, they might do lots of fancy analyses and other things, and many people are then porting their information over into Excel and PowerPoint and making their graphs and charts and various visualizations there.

Does analytics drive business performance?

Bob Thompson:
I want to just ask you about, to me, the most critical business issue: Does this analytics capability in companies really drive company performance?

The reason I ask that is there’s a very interesting chart in your report that shows that those that are more proficient—I’ll just read it off here. Companies with very high corporate analytics sophistication—in your survey, 12 percent of the respondents rated themselves in that category—also rated themselves much better than peers than the rest of the respondents. So, that’s a simple correlation, which you note in your report, and thank you for that. But what conclusion would you draw from this? Is this a driving factor? Is it just a spurious correlation? What role does analytics really have in company performance, in your opinion?

Source: Rexer Analytics 2011 Data Miner Survey

Karl Rexer:
Yeah, well, thank you for noting that it is just correlation analysis. We do try to make sure we make that point, too, because it is self-reported data from people and it’s about their opinions about their company’s performance and their opinions about the analytics sophistication of their company.

We do see that those things seem to be related in how people are reporting those, but I think it does go beyond just how people are reporting them. I think that you really have to look to some work that’s been done by several other people such as Tom Davenport at Babson or Jeanne Harris at Accenture, and they’ve also written books together about competing on analytics. And really I think that their work is a great example of illustrating that the companies that are focusing on analytics are also doing quite well in the marketplace. So, I think that that helps to back up our correlation analysis in our survey.

Bob Thompson:
OK, fair enough. All right, well, do you have any just quick words of advice? I think this media coverage about big data and analytics and all the great tools that are in the market is going to stimulate business people to say, “Hey, I want some of that goodness.” And I know you’ve been in this space for a long time. Can you give just some quick words of advice on how to get going and maybe avoid some of the bigger pitfalls out there?

Karl Rexer:
Well, I just think the biggest thing is not to delay and encourage people to get started. So, like I said, I think it really can be a competitive advantage for many businesses to get involved and analyze either internal data or external data to help them in their making of business decisions. So, I think it’s a mistake to postpone and not get started because your competitors will be.

Now, to get started, I think that executive sponsorship is often critical. So, if you can get high level executives within a company that support this and that support the idea that they’re going to make decisions based on data, not on their sort of gut level feelings about things, that’s important, and to explore and give a chance for some trial and error in the areas of analytics. There might be some mistakes. Everything’s not going to work, but give a few things a try, and I think you’re going to see value out of it.

I think another thing that companies can do is they can—and I’m sorry if this sounds a little bit like an advertisement—but they can get some help for their internal analytic resources. Our company often comes in and helps to get some pilot projects started because it really helps to have some people who have done this before across different industries when you’re getting started. So, if you’re trying to hire people internally to get going, I’d say to try to hire people who have actually done this before, or to make use of companies like my own or other analytic companies that consult to bring in some expertise to help you in those initial projects.

Bob Thompson:
Great. Well, it certainly makes sense. It’s a very complex field. It’s not like you can just pick this up off the bookshelf and start doing it. That’s why I see a lot of PhD’s involved in this field.

Karl Rexer:
Yes, and there’s been some hype around data mining and analytics and everything, and you have to be real careful. Just because you have a tool, it doesn’t mean that you can throw your data in and get your answers out. There’s really a process and a methodology for doing this, and that’s where having someone who’s doing it before can be extremely helpful.

Bob Thompson:
Karl, it’s been a pleasure talking with you and helping shed some light on the area of big data, analytics and data mining and how all that stuff works together. And perhaps more importantly, what companies can get out of it. Thank you very much for your time on Inside Scoop.

Karl Rexer:
It’s been my pleasure. Thank you.