A Very Brief Primer on Analytics - Customer Relationship Management

To help you make the right choices, I’m going to do a very brief primer on analytics. It’s brief because this edition is necessarily more rightbrained than left. We’ll cover analytics types just enough to give you some idea of what they are. We’ll also look at some of the newer tools out there to help you figure out that elusive creature known as the social customer.

Analytic Types
There actually are analytic types. No, not psychologist, psychotherapist, and psychiatrist, though writing this book over the past year plus qualifies me for a few sessions with any one of the three. Those are analyst types. The analytic types are descriptive and predictive.

Descriptive Analytics
This is the analytics for “as is.” It is a historic look at a customer’s behavior organization’s performance, or customer segment’s habits. For example, if you ran a marketing campaign, how effective was it? Have the CSRs been improving their call-to-resolution time? Since you completed the implementation of your SFA system, how has the sales team performed in each city that has it? If there are cities that don’t have it, how do sales compare there? Are the logistics and delivery up to the task, now that sales have increased? These are some of the possible uses of descriptive analytics, also known as operational analytics.

Predictive Analytics
This is the analytics for the “could be if, ”rather than the “to be.” This is where developing models of the possible and scoring the likelihood of achieving something that may be possible, identified down to the individual level, become an important part of the analysis. Predictive analytics take customer data and identify customer segments or individuals and forecast possible behaviors based on historic performance and other factors introduced into a model. They then try to figure out how to utilize the likely outcomes for the benefit of the company. For example, if you reduce the sales team’s administrative time by 12 percent and provide them with the means to get information this much more quickly, what is the possible impact on cost and on your revenue? Or if you promote a specific price cut on a product to 18-year-olds with driver’s licenses in Arkansas, what is the likely increase in responses and in sales based on that promotion? If you remember in previous Chapter, I spoke of Oracle’s Sales Prospector. It uses predictive analytics to see what a deal’s likelihood of successful closing is, what time frame it can optimally close in, and what the likely size of the deal will be.

Once you get past the types of analytics, there are a few analytics toolsets that need to be reviewed because of their importance to business. That would be business intelligence for the quant in you and text analytics for the softer side.

Business Intelligence (BI) Is a Separate Matter
Business intelligence remains one of the most important analytics applications linked to CRM. It was important enough that in the years from 2007 to 2009, every mega-giant business intelligence application provider was snapped up by an even larger company. So we saw Oracle snap up Hyperion, SAP snap up Business Objects, and IBM snap up Cognos. Yum. Why did they make these multi-billion dollar acquisitions? Because it made sense to them to take the most sophisticated BI purveyors with the largest customer portfolios and add their tools and customers to their own offerings. They were and are totally right about that.

To clarify what BI is, I’m going to answer some questions I’ve actually gotten and a few that I made up, thinking they might be important too.

What Is Business Intelligence?
Business intelligence is the use of an organization’s disparate data toprovide meaningful information and analysis to employees, customers, suppliers, and partners for more effective decision making. It is acritical component of a CRM strategy.

Is Business Intelligence Strictly a CRM Thing?
Customer intelligence is not the only BI that exists. Other BI that is frequently found includes product, services, supply chain, financial, and human resources intelligence. In fact, BI extends along the entire enterprise value chain and, even though that value chain is organized around customers, the BI can be broken down to specific links in the chain. The more information you have on each customer interaction throughout all steps of the value chain, the clearer and more innovative thinking you can do on how to treat those customers appropriately, either by segment or, if your information is granular enough, down to the individual.

Isn’t BI the Same as Enterprise Reporting?
In a word, no. Enterprise reporting is the combination of multiple reports from multiple systems using a standard reporting tool and a common delivery platform. It reports the “as is” data that you want it to report from multiple places. It is an impartial information aggregator that grabs already analyzed and interpreted data from multiple sources, and neatly (if it’s working well) ties the data together in a format that makes it readable. It’s not more than that.

The analogy is sort of obvious. When you wrote a paper in college, you handed in a Word document or if you’re as old as . . . my colleagues, a typewritten one. Are the actual physical pages the same as the thought and content? Nope. The physical pages are just the delivery vehicle. When I write a white paper, I usually have a clause in the contract with those hiring me that states that they own the work product, but I maintain my rights to use the ideas elsewhere. If I didn’t, there are parts of this book that couldn’t be written. The book, the white paper, the college term paper are all delivery formats to present the ideas. The ideas are not the reporting of them.

Is BI the Same as Predictive/Descriptive Analytics?
Not exactly. It uses both predictive and descriptive analytics, particularly the former, but isn’t the same thing. The analytics features of business intelligence help present that actionable information and help you make decisions.

Business intelligence takes raw data and turns it into information. It uses complex algorithms on captured data in the “as is” state and makes some interpretive sense out of it. The information, once mapped, interpreted, and identified, is then presented through an enterprise reporting tool in a way that makes it intelligible to those of us with ordinary mathematical skills. That information provides tremendously valuable input for developing the innovations and approaches to improve customer experiences. Some of the general uses of business intelligence, according to the OLAP Report, are:

  • Data warehouse reporting
  • Sales and marketing analysis
  • Planning and forecasting
  • Financial consolidation
  • Statutory reporting
  • Budgeting
  • Profitability analysis

Other possible BI benefits are to help you determine what you are going to sell to whom and when that sale is going to occur. For example, financial services companies might want to see what products will be appropriate for what segments during a certain time of year. You might be interested in creating a scholarship-related instrument that would be sold to 35- to 49-year-olds with children during a school year while their children were between 3 and 15 years old. The BI engine would help you decide whether or not this was a good time of year, a good segment, and/or a good product.

What Are BI’s Challenges?
Implementing BI always has challenges. They fall into two categories, internal and technical. Internally, interdepartmental politics create fragmented, nonintegrated approaches. Simply put, the department wants what benefits it, not what benefits the company. Ease of use is an issue. As far back as 2003, the OLAP 3 Report found that the inability to get users to agree on requirements is a common problem with BI implementations, and if the requirements areagreed upon, staying the course without changing the requirements proves difficult.

There are technical challenges too. The dispersion of the data sources, the “dirtiness” of the data and lack of standards for a common data format, the disparate technologies that are being used, the availability of web services—or not, the sufficiency of the hardware and software to do the job, the sheer size of the total data available—all provide significant challenges to the application of those pesky and complicated analytic algorithms.

For example, First Union has a 27-terabyte database with 16 million customers. Using SAS and Microstrategy, they determined what toup-sell and cross-sell to profitable customers. Technically, that meant 27 terabytes of centralized, normalized, and clean data just to make the data on 16 million customers useful. Then they needed to determine the relevant criteria for the data so the analytics engine could work. The data had to include the account information, sales and purchase data, demographic data, profile data, service/support records, shipping and fulfillment information, campaign responses, and finally, web and other touchpoint data. They did it and it worked, but imagine the effort involved in just readying things.

BI Well Done, Customer Value Received
If you meet the challenges that BI presents in a CRM environment, then the value of the customer intelligence returned can be immeasurable. It helps you identify four basic customer value categories:

  • Customers to retain These are the high value customers that should get the most attention because they will provide the highest profitability. However, as we’ll shortly see, customer lifetime value is an adequate measure by itself.

  • Customers to acquire These are the customers that have high value potential based on their segments and the relevance of the products or services of the offering company.

  • Customers to grow These are the future high value customers that will become the company’s and its partners’ long-term investment. These are strategic accounts.

  • Customers to harvest These are the low value, low margin customers or product/service offerings that can be gathered to the bosom of the company by optimized services or pricing with a minimum of effort and investment. Again, we’ve also discussed in prior chapters the risk associated with just letting these low value customers go because of their ability to socialize things.

Once you have this information, you have to plan to do something with it, not assume it is another notch in a belt or a thing to be catalogued and forgotten. Bells and histles have to be rung and whistled. Use what you have received as valuable customer ntelligence you now can act on. Other than that, it’s just another way to look at data, hardly worth the million or two dollars you spent on it.

I ♥ Text Analytics, a.k.a. Text Mining
Text analysis is to unstructured data as business intelligence is to structured data. What I mean by unstructured data is the conversational data that is going on across the social web. For example, you might find some of it in a comments field beneath a YouTube video while at the same time will find the YouTube video embedded in a blog posting and comments on the video at the end of the blog posting that very same day. But it might be the posting on the wall of Facebook or a threaded forum discussion topic. It might be in the body of an e-mail or in the body of a presentation stored on SlideShare. It could be in a feedback form or from a survey. All in all, the information isn’t exclusive to a single location;it has no standard format that can be easily identified. Most important, the information is freeform, not field delimited, and is without metadata.

But that’s unstructured data. What we need to briefly see is what text analytics does to that unstructured data.

The Definition—of Text Analytics, Not
My favorite definition (by someone other than me)comes from text analytics guru Seth Grimes in his Text Analytics Basics series of articles:
Text analytics:

  • Applies linguistic and/or statistical techniques to extract concepts and patterns that can be applied to categorize and classify documents, audio, video and images.

  • Transforms “unstructured” information into data for application of traditional analysis techniques.

  • Unlocks meaning and relationships in large volumes of information that were previously unprocessable by computer.

What makes it a bit more than just another form of analytics is that most of the nontransactional data that you need to help you improve your individual customer insights and to determine what others are saying about you is unstructured and floating all over the social web.

What also makes this more difficult is that it is the combination of structured and unstructured data into a useful report that gives you the information you need to make intelligent judgments about your customers or on a course of action. Luckily, the tools exist to do this.

How Text Analysis Works
Text analysis, by and large, is relational. A tool extracts information from a source by isolating specific information. Wait. Let’s do it by example.

You use a social media monitoring tool to compile a report that indicates there are 200 or so conversations going on about United Airlines across the Web throughout various channels. The SMM tool compiles the information, but the analysis it does is on the relevance of the piece of information to you, not the analysis of the content of that information. So the SMM tool will separate discussions of United Airlines from discussions of airlines uniting for some legislative action, for example.

When you launch your analytic tools, several actions are taken on the content of the information that you’ve aggregated. Natural language processing (NLP) is applied to structure the content, evaluate the content, extract distinct elements and define the relationships among those elements. Then the attributes associated with those elements are identified and abstracted. An example of this would be the use of sentiment analysis on sentences such as “I can’t stand United Airlines”—you’ll see why shortly. The data is then organized into an understandable and actionable report that helps you judge your strategy toward a group or even down to an individual customer.

If you want a detailed but understandable look at how text analytics works, I would head over to the following online pieces, both by Seth Grimes: “Text Analytics Basics, Part 1” and, you won’t believe this, “Text Analytics Basics, Part 2”. They offer an excellent primer on text analytics, well worth the investment of time.

There are a number of pure play text mining specialist firms like Clarabridge, Attensity, and ClearForest that are worth looking into.

An Easy Case Study: Hewlett-Packard WaterCooler
WaterCooler is a text analysis tool used by Hewlett-Packard that indexes what employees say on internal and external blogs. The capability is there for workers to opt in or opt out. Even with the option to shield themselves from analysis, 11, 000 HP employees have chosen to let all their musings be aggregated. This kind of analysis is more benign than therapy, I suppose. What happens is that the aggregate information then spits out tags that indicate what the 11, 000 employees consider hot at the moment. So, if they are geeking out for the day, or week, you might see “server automation” as a hot topic, though, IMHO, that’s really not very hot. However, if it is hot, that might indicate some course of action to a management team at HP. If the subject that’s being discussed notably is “changeover in HP management team, ”it might indicate some frank discussion or damage control or support needs to go on.

This differs from trends tools such as Twitter Trends, which merely count search keyword instances and then make the tags larger based on the relative frequency with which the words appear. They don’t use NLP or have any sophisticated reporting approach. They pretty much monitor volume only.

Love It or Hate It, It’s Sentiment Analysis
Sentiment analysis is really a particular aspect of text analysis. It finds, evaluates, and processes attitudinal information. Unlike the “information extraction” segment of text analysis, this isn’t just finding facts. It is aimed at customer and market knowledge based on customer behaviors and customer emotions. Do customers love or hate your company? What are they saying and how are they saying it?

If you remember, in early 2009, Domino’s Pizza had a public relations disaster when two employees of a Domino’s restaurant did something disgusting to pizzas that were going to be delivered, then filmed it and uploaded it to YouTube. If you were Domino’s, you would want to know what customers were saying on Twitter, Facebook, in blogs, on the news, and so on. But not just where they said it or what they said as a fact, you’d also want to know what they felt about it and how vehement those feelings were—and how viral was the vehemence— good or bad. (In the case of Domino’s, there was no good, trust me.) That’s where sentiment analysis comes in.

Sentiment analysis measures factors such as expressive words-hate, disgusting, wonderful-words that identify tone, known in the world of sentiment analysis as polarity. It also looks at intensifying words— very, more—words that increase the strength of the tone. It not only looks at the negative or positive (or neutral) nature of a document, but looks at the same opinion state at the sentence level or the opinion associated with an entity—which could mean a word that relates to a human being—a name, a phone number, or an address, for example.

But it’s more complicated than that. University of Illinois professor Bing Liu, a text analytics expert, points out that if a document has multiple entities within it, just deriving the “opinion” of the document isn’t sufficient or even that useful. The multiple entities’opinions have to be ascertained.

If you look at the February 16, 2009, entry in my ZDNet SocialCRM: The Conversation blog called “CRM and the Mac is Like Oranges to, uh Apple, ”you’ll note there are 41 “TalkBacks, ”which means comments in ZDNet-ese. If yougo through them, you’ll find outraged Mac fan boys who call me an “arrogant imbecile, ”something that might be true but only my family’s allowed to call me that. In the case of “arrogant imbecile, ”the entity it’s associated with turns out to be me. The tone is obviously negative. That is sentence-level sentiment analysis. That one is easy.

But what do you do with a comment like this:
First of all, spell out terms like CRM first, before spouting them ad nauseam. Secondly, FileMaker could very easily provide any sort of CRM you would ever need. There are paid consultants available. This is not a high powered database solution by any definition of the term. Finally, learn to spell Mac. It’s NOT an acronym.

On the one hand, there is nothing at the sentence level that is particularly negative. If viewed in that way, it sounds almost helpful. But at the document level, it’s another story entirely. Then it becomes clear that the comment is written by an upset respondent to the blog posting.

So the factors that need to be taken into account here, aside from the respondent’s nasty side, are context, the material that he’s responding to, and the larger tone of the document, which is not that easy to identify as negative. The tip-off terms that would have to be prepopulated for the NLP to do what it’s being paid to do would be words like:

  • “spouting” (which would then have to be abstracted from whales)
  • “NOT” in capital letters
  • “ad nauseam”

None of them an obvious barn buster, but all together they indicate the document (comment) tone.

That comment is a perfect example of the difficulty of sentiment analysis. Yet, this is the moment in the history of business when text analysis in general and sentiment analysis in particular are becoming critical. The conversation needs to be heard and the data it provides needs to be captured if you hope to have any success with your customers.

Many companies are adding sentiment analysis to their products. In mid-2009, megamonster company SAP announced a product that uses sentiment analysis. SAP announced a customer service application that scores sentiment in Twitter feeds related to customer service. Using the Business Objects Insight product, they analyze Twitter feeds that indicate customer issues and the emotional intensity of the feeds. The product then looks at the potential for a viral outburst given the conversations going on around it. There are business rules that identify courses of action when the resulting sentiment scores come back in the context of what the problem is. Those rules are used to trigger an alert that gets sent to the parties who are given the responsibility to deal with problems of a particular type and intensity.

This isn’t unique to enterprise companies either. Social media monitoring companies are now adding sentiment analysis so that they cannot only track the relevance and frequency of whatever it is you want to track, but also the level of friendly or hostile activity.

Radian6, added sentiment analysis to their product in late 2009 so that it not only aggregates the conversations around the subjects being listened to in real time, but also scores them for their emotion. This adds an entire dimension to listening. Radian6 uses the product to monitor anything ranging from customer service issues to consumer opinions on products. One use of the product is to see how the user of a company’s services or products solves a problem. Rather than scouring forums for the results, the use of sentiment analysis can see how positively or negatively the community responded to the solution. If the solution is good and responded to with a thumbs up from the user community, then the company can integrate it into their own knowledge base. All of this is aggregated into a dashboard that the Radian6 user can see.

Structured + Unstructured = Home Run
Text analysis is one of the most important components of analytics introduced into the mix of available products. But the real power lies in integrating the unstructured data from the social web with the structured data inherent in operational CRM systems. While I’ve spoken frequently about integrating social media monitoring into CRM systems, this is another step entirely and far more powerful because of the rich customer knowledge it provides. The combination of transactions and conversation with personal details that have been harvested from the Web gives companies a very valuable look at either an individual customer or a trend that is moving in real time and its possible impact on your company.

The combination of historical record and real-time interaction is not without its problems. For structured and unstructured data integration, there has to be a default to some sort of structure so that the data can be combined in ways that are beneficial to the business.

Using XML as a bridge between the two data types is already a mainstream approach to this knotty issue. The idea is to take unstructured data, extract it, and then incorporate it into a standardized XML format. Structured data can already use an XML format, so XML becomes the common denominator for the integration of the two types of data.

The XML tags or the database table schema are predefined by the users. Then technology can be applied to tag the unstructured text. That can be a dictionary look-up, some sort of machine learning, or a rule-based pattern-matching application. These technologies identify and tag domain-specific information. Once all the data is “entered” into the XML schema, it needs to be cleaned and readied for whatever future reporting needs or slicing and dicing need to be done to create knowledge for insight from the data. The resultant XML file is in a form amenable to query, search, and integration with other structured data sources. Then all the data is imported to some centralized data store.

The value? You can see interactions and transactions of a single customer over time in ways that actually allow you to improve how you interact with that customer. Not only that, but since study after study shows roughly 80 percent of all data that an enterprise has access to is unstructured, the integration becomes invaluable to searching and finding the data to be used. Slick stuff.

The Future of Analytics
I have to admit that analytics isn’t necessarily the most exciting part of CRM to me—though sentiment analysis is kind of cool. As it’s constituted now, it’s a way of finding, capturing, and (duh) analyzing data so that there’s organized useful information that becomes an important feature in making business judgments. Hey, if that’s exciting to you—you know the old expression—whatever floats your boat.

But there is research going on with vendors and technology mavens right now that actually could make analytics not just utilitarian. This is the use case that you might be considering as a real story in just a very few months or maybe a year from when you read this (now that’s a real trick . . .).

The Scenario: Old School
As things go now, you know that when you call customer service, most likely, a screen will pop up on the customer service representative’s (CSR) side. The CSR will see what your history has been. If complete, that history will have your past purchases, your past complaints and their resolution, and any marketing data (such as campaign responses or literature requests). It will have what kind of contract or SLA you have with them and flag the appropriate level of service that you should be provided. It will identify if you are a customer who has to be prioritized. It will show what forms of media you use most frequently to interact with the company (e-mail, phone, etc.). If it’s a sophisticated system, the CSR will have the ability to see how the problems had been resolved in the past and the links/screens with the potential solutions in front of them, given the probability of the same issues occurring. He or she will be able to quickly call up the possible answers to the problem due to an extensive knowledge base. The workflow is based on the SLA and scoring you as a customer based on your transactions, maybe even customer lifetime value or other metrics, to see how “important” you are to the company. The number and level of complaints in the past will drive whom you speak with, how quickly it escalates, and so forth. The trouble ticket is assigned and the dance begins.

The Scenario: New School
It starts the same way. You call customer service about problems with the company’s product. This time, when your call is in and the complaint is generated, your customer record, the appropriate SLA level, and the potential solutions from the CSR’s in-house knowledge base pop up. But now, information that has been captured from the unstructured social discussions about the product and customersuggested solutions, and your conversations on Twitter, Facebook, and the Consumerist appear like lightning. The content of the unstructured information is automatically analyzed to see if it’s about the same subject that you’re calling about, and what your past relationship to the company is. Sentiment analysis is done on the spot on your external conversations to help the CSR determine how severe the issue is, what kind of complainant you are (chronic, one time, valid, whiner), whether you are an influencer, and so on. Based on the analysis, the CSR knows how to treat this case—a.k.a. you. It might mean sending you to a supervisor right away or working with you at a level that, in a prior era, couldn’t have been determined.

If the problem is not solvable on the phone or whatever medium it’s being addressed through, when the ticket is assigned, it’s not just a specialist who gets the problem to solve, but a combination (potentially) of internal and external communities. Once they have used their combined intelligence to come up with an answer, then the answer is refined, enriched, brought in-house, and added to the knowledge base while your customer ticket is being punched as complete.

On a more disconnected scale, this is already going on. Companies like Samsung and Procter and Gamble are using external social networks to capture customer data and improve their product knowledge about everything. Remember previous Chapter’s “Superstah! Helpstream” found that their clients who have customer communities get 17 percent of their customer service issues taken care of by the customers themselves. IT issues are often solved elegantly in the thousands of independent technical threaded discussion forums without any support of the company involved at all.

What isn’t common is the aggregation of the unstructured external data, meaning the data extracted from the conversations going on in forums outside the company in a form that’s usable by customer reps anywhere. However, the means of capturing structured and unstructured data into a useful report is becoming more easily available with companies like SAP and Oracle on the mega-side to much smaller companies like open source BI provider Omniture making them available through their applications.

Customer insights don’t come free, but they do come if you interpret the data and make smart judgments.


All rights reserved © 2018 Wisdom IT Services India Pvt. Ltd DMCA.com Protection Status

Customer Relationship Management Topics