67. 정확도
285
ACCURACYORPRECISION?
In an ideal world, you want to be both precise and accurate. The difference
between the two ideas is illustrated using a bull’s-eye target in Figure 10.1. In the first
target, the hits are closer to the bull’s-eye and have a higher degree of accuracy. But
each hit is quite spread out. In the second target, the hits are close to each other, quite
precise, but not accurate.
High Accuracy,Low Precision High Precision,Low Accuracy
Image Source:Wikimedia Commons
Figure 10.1 The difference between accuracy and precision
So, which approach should you choose as your strategy for collecting web
analytics data? Precision. The reason is simple: precision is predictable and therefore
reproducible; you can be significantly more confident taking action with insights from
정밀도를
78. 방법
A Six-Step Process for Dealing with Data Quality
The Web is still a baby. It is going to grow and evolve, in bursts I am sure, but the cur-
rent methods for collecting data are imperfect. That will change over time, but for now
we have to face the facts and figure out how to most effectively optimize our strategy.
We have already agreed to deal with measurement using the precision men-
tal model. Over time you can become more accurate about collecting precise data.
Figure 10.2 shows a six-step process I have developed so you can implement a virtuous
data quality cycle in your company.
Move Fast,
Think Smart
Clean Data
Collection
Be Very
Selective
Incomplete?
No Problem.
Go for
Precision
Data
Audits
Figure 10.2 The virtuous data quality cycle
Here is more detail on each step of the process:
1. Collect clean data: Follow all the best practices to collect your data, and don’t
•Clean
282. 이동
96
ALYTICSTRAPS
Articles in media,
conversations with
friends
Product
Research and
Evaluation
Offline Behavior
Searching websites
for reviews and
discussion
Visiting branches and
gathering sales
literature
Supplier and
Product
Selection
Search by company
and product on price
comparison/review
sites
Pre-purchase
Qualification
Comparing price,
delivery,and support
across several
possible sites
Buy in-branch or order
over the phone and
complete by post/in
branch
Product
Purchase
Payment or commit
to purchase online
and complete by
post/in branch
Online Behavior
Talking to staff,or
asking questions by
phone
Figure 10.7 Nonline customer behavior scenarios
Nonline marketing goes against the grain of everything we have been taught so
far; it messes with our comfort level. But in a Web Analytics 2.0 world, you must strive
다중
370. 도전(2/3)
300
10:OPTIMALSOLUTIONSFORHIDDENWEBANALYTICSTRAPS
amounts of data, at a level that is difficult for a human, or even our analytics tools, to
parse it, interpret it, and identify insights fast enough. Due to our own limitations, that
of our organization structures, and our decision-making layers, we struggle to take
action even every few days. This is why we can rarely extract even marginal value from
getting real-time data.
Diversity Most website owners are pretty bad at understanding all the reasons why peo-
ple come to their websites (see Chapter 6 for how to overcome this challenge). People
may use your website for purposes that you did not intend, and that complicates data
analysis and identifying action.
When you do behavior targeting with the right tool, you can overcome the scale,
data interpretation, and diversity problems by automatically understanding your visi-
tors as they interact with your web presence and showing them the most relevant con-
tent. Figure 10.11 shows a schematic of a typical on-site behavior-targeting system.
Visitor Arrives at Your Website CMS (Serves Content)
Call Goes Out to
Visitor Profile
Repository
Build Profile
Visitor Profile
Repository Automated or Rules-Based
Modeling Engine
Optimal
Content Decision
Sent to CMS
Content
Library
First-Time Visitor
Retrieve Profile
Repeat Visitor
Figure 10.11 How automated behavior targeting works
The top half of Figure 10.11 shows what’s happening on your website today. The
bottom half is what happens when you use behavior-targeting tools.
You can base your content targeting on business rules that you input into the
• 일반적인
548. 단계
307
PATHTONIRVANA:STEPSTOWARDINTELL
JavaScript
Tag
30
Page Views
Visitors
Time on site
Referrers
Configure
Settings
The
LadderofPain
to
N
irvana
10
Internal
Site Search
Goals
Funnels
Campaign
Tracking
70
Email
Affiliate
Paid Search
TV/Radio
Revenue
60
Cash!
Custom
Variables
Metadata
Rich
Media
85
Flash
Video
Web Apps
Widgets
Figure 10.12 Optimal path to successful web analytics evolution
There are five simple steps to analytics nirvana. Each step makes incremental
progress, each allows you to get more data and make better decisions, each has associ-
ated pain, and each step attempts to balance cost and benefit. Let’s cover the logic and
•1