Data mining for business decisions

Please read teh case and answer 3 questions:

CASE:Applebee’s, Travelocity, and Others: Data Mining for Business Decisions.

Randall Parman, database architect at restaurant chain

Applebee’s International and head of Teradata’s user

group, opened Teradata’s annual user conference in

Las Vegas with a warning to those who aren’t making the best

use of their data. “Data are like gold,” Parman noted. “If you

don’t use the gold, you will have someone else who will

come along and take the opportunity,” speaking to a room

packed with almost 3,900 attendees.

Parman drew an analogy to the story about Isaac

Newton’s discovery of gravity after he was hit on the head

with an apple. “What if Newton had just eaten the apple?” he

asked. “What if we failed to use the technology available, or

failed to use these insights to take action?” Applebee’s, which

has 1,900 casual dining restaurants worldwide and grossed

$1.34 billion in revenue last year, has a four-node, 4-terabyte

data warehouse system. Although the company has a staff of

only three database administrators working with the system,

“we have leveraged our information to gain insight into the

business,” he said. “Some of those insights were unexpected,

coming out of the blue while we were looking in a completely

different direction.”

For example, Applebee’s had been using the data warehouse

to analyze the “back-of-house performance” of restaurants,

including how long it took employees to prepare food

in the kitchens. “Someone had the unanticipated insight to

use back-of-house performance to gauge front-of-house

performance,” he said. “From looking at the time the order

was placed to when it was paid for by credit card and subtracting

preparation meal time, we could figure out how

long servers were spending time with customers.” Parman

added that the information is being used to help the company

improve customer experiences.

Applebee’s has also advanced beyond basic business decisions

based on data—such as replenishing food supplies according

to how much finished product was sold daily—to

developing more sophisticated analyses. His department, for

example, came up with a “menu optimization quadrant” that

looks at how well items are selling so that the company can

make better decisions about not only what to order, but

about what products to promote.

Meanwhile, technology vendors see untapped potential

for businesses to spend money on software and hardware

that lets them use data to make more sophisticated

business decisions. “Companies who operate with the

greatest speed and intelligence will win,” says Teradata

CEO Michael Koehler.

Like many companies, has lots of unstructured

data contained in e-mails from customers, call

center representative notes, and other sources that contain

critical nuggets of information about how customers feel

about the travel site. To offset the inability of business intelligence

tools to search for unstructured data, Travelocity has

launched a new project to help it mine almost 600,000

unstructured comments so that it can better monitor and respond

to customer service issues.

The online travel site has begun to install new text analytics

software that will be used to scour some 40,000 verbatim

comments from customer satisfaction surveys, 40,000

e-mails from customers, and 500,000 interactions with the

call center that result in comments to surface potential customer

service issues. “The truth is that it is very laborious

and extremely expensive to go through all that verbatim customer

feedback to try to extract the information we need to

have to make business decisions,” notes Don Hill. Travelocity’s

director of customer advocacy.

“The text mining capability . . . gives us the ability to go

through all that verbatim feedback from customers and extract

meaningful information. We get information on the

nature of the comments and if the comments are positive

or negative.”

Travelocity will use text analytics software from Attensity

to automatically identify facts, opinions, requests, trends, and

trouble spots from the unstructured data. Travelocity will

then link that analysis with structured data from its Teradata

data warehouse so the company can identify trends. “We get

to take unstructured data and put it into structured data so we

can track trends over time,” adds Hill. “We can know the frequency

of customer comments on issue ‘x’ and if comments

on that topic are going up, going down, or staying the same.”

Unlike other text analytics technology, which requires

manual tagging, sorting, and classifying of terms before

analysis of unstructured data, Attensity’s technology has a

natural language engine that automatically pulls out important

data without a lot of predefining terms, notes Michelle

de Haaff, vice president of marketing at the vendor. This allows

companies to have an early warning system to tackle

issues that need to be addressed, she added.

VistaPrint Ltd., an online retailer based in Lexington,

Massachusetts, which provides graphic design services and

custom-printed products, has boosted its customer conversion

rate with Web analytics technology that drills down

into the most minute details about the 22,000 transactions it

processes daily at 18 Web sites.

Like many companies that have invested heavily in online

sales, VistaPrint found itself drowning, more than a year

ago, in Web log data tracked from its online operations.

Analyzing online customer behavior and how a new feature

might affect that behavior is important, but the retrieval and

analysis of those data were taking hours or even days using

an old custom-built application, says Dan Malone, senior

manager of business intelligence at VistaPrint.

“It wasn’t sustainable, and it wasn’t scalable,” Malone

says. “We realized that improving conversion rates by even a

few percentage points can have a big impact on the bottom

line.” So VistaPrint set out to find a Web analytics package

that could test new user interfaces to see whether they could

increase conversion rates (the percentage of online visitors

who become customers), find out why visitors left the site,

and determine the exact point where users were dropping off.

The search first identified two vendor camps. One group

offered tools that analyzed all available data, without any upfront

aggregation. The other offered tools that aggregated

everything upfront but required users to foresee all the queries

they wanted to run, Malone says. “If you have a question

that falls outside the set of questions you aggregated the data

for, you have to reprocess the entire data set.”

The company finally turned to a third option, selecting

the Visual Site application from Visual Sciences Inc. Visual

Site uses a sampling method, which means VistaPrint can

still query the detailed data. but “it is also fast because you’re

getting responses as soon as you ask a question. It queries

through 1% of the data you have, and based on that . . . it

gives you an answer back. It assumes the rest of the 99% [of

the data] looks like that. Because the data has been randomized,

that is a valid assumption,” notes Malone.

VistaPrint, which has been using the tool for just over a

year, runs it alongside the 30–40 new features it tests every

three weeks. For example, the company was testing a fourpage

path for a user to upload data to be printed on a business

card. The test showed that the new upload path had the

same conversion rate as the control version. “We were a little

disappointed because we put in a lot of time to improve

this flow,” he adds.

When the company added Visual Site to the operation,

it found that although the test version was better than the

control in three out of four pages, the last page had a big

drop-off rate. “We were able to tell the usability team

where the problem was,” Malone says. VistaPrint also reduced

the drop-offs from its sign-in page after the Visual

Site tool showed that returning customers were using the

new customer-registration process and getting an error notice.

The company fixed the problem, and “the sign-in rate

improved significantly and led to higher conversions,” he

says. While Malone concedes that it is hard to measure an

exact return on the investment, the company estimates that

the tool paid for itself several months after installation.


What are the business benefits of taking the time and

effort required to create and operate data warehouses

such as those described in the case? Do you see any

disadvantages? Is there any reason that all companies

shouldn’t use data warehousing technology?


Applebee’s noted some of the unexpected insights obtained

from analyzing data about “back-of-house” performance.

Using your knowledge of how a restaurant

works, what other interesting questions would you suggest

to the company? Provide several specific examples.


Data mining and warehousing technologies use data

about past events to inform better decision making in

the future. Do you believe this stifles innovative thinking,

causing companies to become too constrained by

the data they are already collecting to think about unexplored

opportunities? Compare and contrast both viewpoints in your answer.