3 Main Types of Data Analysis used in CRM

This article throws light upon the three main types of data analysis used in CRM. The types are: 1. OLAP (Online Analytical Processing) 2. Click Stream Analysis 3. Personalization and Collaborative Filtering.

Data Analysis: Type # 1. OLAP (Online Analytical Processing):

Despite its varied interpretations, ‘data mining’ has acquired an almost mystical allure over the past decade, although its widespread interpretation is as an activity associated with querying increasingly detail data -“drill down,” as its called – in-fact, data mining is a heavily specialized sub categories of analysis that has specific application from both with in outside CRM.

In-fact, the term drill down is more appropriately applied to the practice of online analytical processing, known as OLAR. OLAP has become the most popular type of decision support analysis, allowing the average business person to explorer data online with the aim of focusing on data at a lower and lower level of the data hierarchy. Most often, this mean generating an online report, analyzing the result, and submitting a more detailed query in order to understand the result data.

OLAP generally focus on proving a set data attribute from a data base organized around certain dimension, such as time and location, thus a user can request the companies regional sells revenue for all baby care products by region or by store. He can request the report detailing regional revenue for each month with in quarter.

Although OLAP is generally lumped into the data mining rubric-usually by software vendors eager to claim the data mining moniker-it normally relies on data that has been summarized according to particular dimension, data mining involves the identification of meaning full pattern an rules from detail data, usually from large amount of data, thus, instead of analysis customer segment to determine who is likely to churm as with OLAP, data mining would examine individual customers , touching each of the millions of records in a database.

OLAP analysis required the analysis to have a query or hypothesis in mind, but data mining can generate information to show pattern and relationship without the analysis knowing about them. Data mining can identified cluster of customers who buy similar products.

For instance, home office workers who buy PCs , power supplies, toner, printer cables, waste papers baskets, and coffee, with an OLAP tools, the anlaysist would have to guesses which product a home office workers would purchase and then identified customers making such a purchase. OLAP analysis typically examine category grouping such as PCs, printer cable, and toner [computer related products] but might not organized out of category purchase such as coffee and waste baskets.

Where theory meets practice -Data mining in CRM:

Data mining tool identified pattern in data and deliver valuable new information that can increase a company understanding of itself an its customers. Data mining is commonly used to help data analysts search for information they don’t yet no to look for, often involving no hypothesis. It has helped companies uncover a diverse set of new knowledge, from a customer next purchase to optimal store layouts to the most favorable release data for a movie in preproduction.

There are many different types of data mining algorithms, some esoteric and not easily applicable to business problems (multivariate adaptive regression splines, anyone?).

Although the specific algorithms themselves might vary- decision trees and neural networks are fundamentally different but can both we used to predict behavior the following three types of data mining particularly germane to CRM:

(1) Prediction:

Prediction the use of historical data to determine future behaviors. Predictive modeling generates output that populates of “models” or structure to represent the results. For instance, a predictive model can indicate the next product a customer is most likely to purchase based on historical purchases buy that customer and other customer who have purchased the same product.

(2) Sequence:

sequential analysis identifies combination of activities that occur in a particular order . business use sequential analysis to determine whether customers are doing thing in a particular order. It can help a business distill behavior from events captured from various operational systems around a company to determine patterns. For instance, a bank or telecom companies can learn more about a given customer or customer segment by exam­ining patterns in the slowdown of purchases or in service cancellation.

(3) Association:

Association analysis detects groups of similar item and events. It can we used to detect items or events that occur together. The association algorithm is often applied to market – basket analysis to help business understand products being purchase together. By understanding customer and product affinities, a company can make important deci­sions about which products to advertise or discount and which customers should be tar­geted of certain product.

One central difference between data mining and other types of decision support analysis is that data mining usually involves statisticians or product specialists intimate with the use of the correct algorithm and their application to business problem, as well as with the specific data mining software.

Although the business persons rarely mines the data herself, she might use data mining results- either represented graphically in a visualization tool or deployed to a data base for general query access to help make important decision about managing customer relations.

There are myriad uses for the three types of data mining just describe, from targeting brand new customers by modeling existing customer’s response pattern to avoiding high risk prospects through risk prediction or forecasting a customer’s life time values. Many companies acquire dedicating data mining servers, on to which they load customer data record to build models and explore various customer behavior patterns such activities are usually processing- intensive so standalone data mining plate forms avoid impacting processing on other systems.

These servers are usually linked to a company’s data warehouse, enabling data analysts to easily access customer data to experiment with various pricing plans, for example, or to create dynamic customer segments for testing new campaigns and performing what-if analysis.

Each type of data mining can DVD player commutes to work on the train, causing the retailer to reallocate much of its marketing budget from daytime television commercials to newspapers ads and billboards. The company saw sales of these players shoot up 43 present after changing its ad media.

Understanding the impending behaviors of customers and prospects is the key to data mining, and where CRM is concerned, two data mining applications in particular stand out: click stream analysis and personalization.

Data Analysis: Type # 2. Click Stream Analysis:

IT department have become giddy over capturing click stream-the data that illustrates a Web visitor’s footprint around the site, how long he stayed, what he did during his visit, and when he returned. They’re the equivalent of a camera in a department store recording a shopper’s every move.

Click stream data-usually stored either as a part of a company’s data warehouse or in a dedicated click stream data store sometimes called a “data warehouse”-is growing hand-in- hand with corporate e-commerce activities.

One client of mine, a general merchandise retailer who has joined the e-tailing ranks, want its Web site to be as “sticky “as possible and has begun analyzing click stream data to surmise why customers might leave the site prematurely. The company has sharpened its analysis to determine the value of abandoned shopping carts. When a customer leaves the site in the midst of a shopping trip, whatever the reason, the company looks to see what products were in the cart. The data is then compared with similar data from other abandoned carts to examine.

1. How much revenue the abandoned cart represented {in other words, how much revenue was lost because of the customer’s early departure}

2. Whether the products in the cart were high-profit items or loss leaders

3. If the same product were found in other abandoned carts

4. The volume of products and the number of different product categories in the cart

5. Whether the total bill for the abandoned carts consistently fell within a certain dollar range

6. At what point during the shopping trip the cart was actually abandoned {when the cus­tomer saw the shopping charge. When the site required a personal survey before confirm­ing the purchase,

7. How the average and total bills for abandoned carts compared with “un-abandoned” carts- those that made it through the checkout process

The result of this analysis can trigger some interesting theories. For instance, perhaps none of the products in the cart was appealing enough to a particular customer to motivate her to continue shopping or the customer was put off by frequent inquiries asking her whether she was ready to check out or possibly, at a particular dollar total, the customer thought the better of the entire shopping trip and bailed.

Finally, perhaps the number or mix of products in the cart reminded the customer of another site that offered a steeper discount for similar purchases.

Admittedly, some of these theories are mere guesses. But when examined regularly and with consistent metrics, click stream can reveal some interesting patterns. The fact is, whatever the customer’s reason for leaving the site and a cart full of merchandise, the e-tailor can take a verity of actions based on both hard findings and less-than-certain extrapolations the e-tailor can use these results to tweak the design and contents of its Web site and monitor resulting improvements.

Pattern might indicate product affinities, suggesting cross selling and up selling strategies. And when combined with customer’s demographic, psychographics, and past behaviors, click stream data can bring the understanding of customer behavior to a whole new level.

The latter option is perhaps the most intriguing: rather than simply examining a customer’s navigation patterns and guessing about which actions to take, the retailer can navigation patterns and guessing about which actions to take, the retailer can combine those patterns with more specific customer data-his previous purchases in that product category, key demographic and psychographic data, or his lifetime value score, for example-to provide a holistic view of that customer’s value and interest.

It might have been a one-time-only shopper who was lost, but in other cases a high-value customer might have left the sit on multiple occasions. A tailored e-mail message electronic coupon perhaps target one of the products left behind on a prior trip-could make all the difference the next time that customer decides to log on.

The following scenario, based on a real-life case studies illustrates how click stream data, when integrated with other key data from around the enterprise, enhances opportunities to personalized customer communications. Most marketing managers won’t be looking our analyst, shoulders individual click streams. But understanding our customer’s navigation around the site can help a company decide how to lure him back.

You have several chooses .your companies usually tactic for all registered visitors who visit the site but don’t make a purchase it is to mail them to coupon for $5 off a new pair of fashion eye wear. However, this particular visitor was looking at contacts. He’d probably trash the glasses coupon as soon as it arrived in the mail.

A better choice might be to e-mail the visitor a discount code-a coupon is given a unique code so no one but the given customer can redeem it- for $10 off a new pair of hard contacts or three pairs of disposable lenses {a predictive model could confirm this as the best course of action} your profit on contact lenses is usually good, and the shopper seemed on the brink of making a purchase .besides, e-mailing the offer is a lower cost option that the U.S. postal service and would probably result in quicker turnaround time.

Along with this more personalized tactics, you could also monitor the referring Web site for other referred shoppers who have researched or purchased connect lenses, if connect lens activity is particularly high, you might concerned placing a more customized banner ad on the partner’s sit and even provide better financial incentives for the partner when new connect lens customers click through.

With the e-mail strategy, the customer more likely to return to sit and you’re almost guaranteed purchases. WIN-win? Right-wrong.

The problem with this scenario is that even through analysis is involved, it’s still dangers. The fact that you’re looking at only a single customer touch point can mean his problems and a bad decisions. If your click stream database contained behavior history on this shopper, things might turn out differently. You would have more information about the customer, and you’d know the following:-

1. This isn’t the customer’s first visit to your Web site.

2. He has made three other purchases on three separate occasions

3. The products he has purchased have all been on sale

In short, you would understand that you Web visitor is what’s known as a “cherry picker” someone whose only purchase low-margin products when they’re being promoted. No cross-selling, no up-selling, no true loyalty. He’ll be back again, too, when he finds the next markdown.

If you had this information, you would understand the optimal marketing tactic for this customer: do nothing: do nothing, any further marketing to him would be a bad investment. Of course, you’re perfectly happy to have this customer return to your site of his own free will.

But you’ve already invested too much money in an unprofitable. Each times retailer price-subsidizes product for cherry pickers, it is losing an opportunity to sell that product it a more valuable customer. The retailer is in fact investing in an undesirable customer relationship.

Data Analysis: Type # 3. Personalization and Collaborative Filtering:

The practice of tailoring communications directly to a customer segment or, increasingly, to an individual customer. The premise of personalization is that, by collecting sufficient customer data, a company can market to an individual’s unique needs, both now and in the future.

Personalized communications is the principal techniques via which companies can convince customers they understand them and that their information-which the company often uses thanks to the customer’s explicit permission-is mutually beneficial.

The goal is to deliver accurate product recommendations; content geared to individual preferences, and targeted promotions for individual Web visitors-and in real time. When done right, personalized means not only maintaining customer loyalty, but also driving purchases higher leverages detailed information about individual and can dictate some very tactical decisions.

The following analysis topics from a drugstore e-tailor suggest the level of individual detail and resulting tactics personalization can provide:

1. For people who have bought or expressed interest in vitamin supplements, which other products are likely to buy?

2. How likely is customer X to buy prescription drugs online?

3. What other items are likely to be in a shopper’s market basket if he buys, say, decongestant?

4. Which products are most similar to brand eye drops the customer chose?

Personalization can take various forms. It can involve customizing actual Web pages, including a Web site’s look and feel, according to the features favored by an individual visitor. Many Web sites allow the visitor to customize the site according to her preferences, eliminating format variations and allowing her a private window into the company. Use the search function often? Move the search window to the top of the page. Like customization, so called localization can focus site content to the visitor’s particular geographic area.

Notice that the personalization leverage established rules that dictate, for instance, which products might be purchased might be purchased together or whether a certain Web pages should precede or follow another.

When a visitor to a software Web site buys Quicken, the site might suggest he buy Quicken: The Official Guide before going to the checkout screen. Rules- based personalization most often involves rules that have been hard-coded into the software. For this reason, it’s often difficult to maintain and support.

The other type of personalization, adaptive personalization, learns as it goes. More commonly known as collaborative filtering, this type of personalization gats smarter as it observes customer behaviors and applies them to new circumstances. For instance, if a gardening e-tailor using collaborative filtering observes that shoppers tend to buy low-cost perennial flowers at the same time they order gardening tools, the Web site might begin suggesting a flat of pansies to all customers who buy bulb planters.

Collaborative filtering uses the behavior of other “like” visitors as the basis for its recommendations. Collaborative filtering tools are often more complex, and thus more expensive, than rules-based personalization. The most celebrated example of collaborative filtering is Amazon(dot)com’s purchase circles, in which Amazon factors in the buyer’s past purchase and geography to suggest what readers who live in her neighborhood and have similar interests might be reading.

The more similar shoppers buy, the smarter Amazon becomes about their preferences, and the more accurate are the site’s recommendations. Several Amazon(dot)com customers I know are cherry pickers on other booksellers’ Web sites during special promotions, but they always return to Amazon because “they know me better.”

Perhaps the most telling delineation in personalization is in whether or not the user knows its happening. In the permission marketing scenario Web visitor voluntarily provide personal information to Web sites where they believe there will be some sort of Quid pro quo: the company will use the information to provide a value added service such as periodic discounts or special- interest newsletters. Some sites can personalize content without making the shopper aware that the products he’s seeing are different from those of fellow shoppers-who might have different profiles and

Web retailers who combine eCRM with detail customer data and advanced personalization can customize content and screen layouts for individual visitors to increase the site’s stickiness and the shopper’s propensity to buy.

On the other hand, companies such as Lands’ End simply ask customers what they like, whether or not they make a purchase. The company’s My Personal Shopper feature shows Web visitors various product combinations and solicits their feedback.

This practices is different from the “inferential” personalization in which a company applies complex logic to infer a customer’s preferences-“referential” personalization simply stores a customer’s responses to questions or surveys, making those answer part of her profile so they can be used to cross-sell her additional products.

Although custom content seems innocuous enough-it tantamount to reorganizing a brick and mortar store’s layout according to the way the shopper likes to move around the store-it can also have more controversial uses.

Amazon(dot)com was revealed to be selling the same DVD movie for different prices to different shoppers/this practices, known as dynamic pricing, turned the concept of consumer choice on its head. The Web, famous for offering shoppers the opportunity to find the best deal with a simple mouse click, was now allowing sellers the opportunity to differentiate consumers and their price sensitivity.

Dynamic pricing actually leverage CRM technology and detailed customer data to let a company, say, compare a shopper’s desire for the product with his perceived ability to pay for that product. For the first time, consumers are the ones competing for the best deal.

Arguments for and against dynamic pricing raise issues of consumer privacy as well as goodwill.’ after all the more a shopper buys on a company Web sites, the more information the site has on the buyer and the weaker the buyer’s negotiating power.

In the past several years, Airlines were routinely accused of raising tree online fares for frequently fliers-there most loyal customers-who are more likely to fly a particular carrier because of the mileage perks. And, in a now in famous public relations gaff, coco cola was alleged to have been considering a vending machine that raised the price of beverages when the temperature soared in.

In defending dynamic pricing, e-tailor point to their brick-and mortar counterparts who have been engaging in the practice for years. Drug stores have been known to price cold medicines higher in chillier climates, and the short age of Sony play station to units last Christmas trove price up sharply/Kmart used the short age as a way of rewarding loyal customers first, steering play station availability to loyal shoppers on the company’s blue light(dot)com Web site, “in the Web world, where consumer data can include a shopper’s home addresses, income level, number of children, and even his resolve to purchase a product, dynamic pricing-along with a number of other personalization techniques-can be implemented more quickly and to the wider number of the shoppers.

Good or bad, Amazon’s dynamic pricing experiment might have gone unnoticed all together if it hadn’t been for the Web. in an ironic twist participants in an internet chat room begins comparing their movie receipts and discover that price seemed higher for the regular customer.

Amazon claimed the dynamic pricing was simply are test and denied plans to for mailed the practice. But the example proofs the Web has affected both business and social communication to the point where even CRM can sometimes be are double-edged sword.