Brand Website Activity
Impact Analysis:
Overview: A Paradox
An argument that produces inconsistency
•
Search optimization is designed primarily to drive traffic
to a brand website
•
Marketing mix models in the pharmaceutical industry
typically show a strong, positive relationship between
investment in search optimization and resulting Rx or
sales outcomes … “
digital is ROI-positive
”
•
However – most brand website data also shows only a
weak relationship between website visits or page views
and Rx or sales
•
Why this paradoxical result? Our team set out to
discover the differences between types of web visits to
add new insights into the evaluation of search
Brand Website Analytics:
Data-Rich but Causality Uncertain
•
A wealth of web activity metrics:
–
# of visitors to website, by day, by zipcode
–
# of page views per visit
–
Pages viewed / duration of view
–
Sequence or path of page views, by visit
–
Activity or registrations per visit
•
And insights into source of web traffic:
–
Referring source to brand website
–
Keyword search term used to get to site
–
Clicks / Click-Through Rate / Cost-per-Click
Great data
for
optimizing
volume of
web traffic
and
engagement
with the
brand site
…But does it
causally
impact
brand Rx?
Web Analytics Processes
Typically Focused on Optimizing Web Traffic
Paid Media
Website/
Landing Page
Conversion
Optimize
Traffic to Site
Impressions
Clicks
Click-Thru-Rate
Visits
Page Views
Path Analysis
Marketing Mix:
Optimize Paid
Media ROI
Page Views and Rx
Typically Low Predictive Relationship
Paid Media
Website/
Landing Page
Conversion
Impressions
Clicks
Click-Thru-Rate
Visits
Page Views
Path Analysis
Why Not Just Measure Here and
Then Optimize Traffic to Site Via Paid Search?
Typically, low
correlation observed
between page views or
•
Path analysis loses analytic power quickly due to number
of possible paths:
1,956 paths alone for a 6-page website
•
Analytics proved that content viewed was more relevant than
order of page views
AND more relevant than
quantity of pages
viewed
Digital Analytics Case Study
SAI Methodology
Linking “Content Viewed” to Rx Outcomes
Develop Site-Specific
SAI Metric
Collect Data
Create Model
1) Categorize Site
6) Quantify Impact on Sales (Rx) Over
Time
3) Score Each Site Visit
4) Aggregate Scores by Geography
5) Aggregate Sales, Market Events, Other
Promotions by Geography
Steps 1-2: Scoring Methodology
High-level Site Audit
1) Categorize Site: Understand drivers influencing key site metrics
Using unique visit-level
data from web tracking
tool, identify the specific
site assets used each
session
Identify the order of
events
Track whether call to
action is performed
Create profiles based on
site activity
1
2
Steps 1-2: Scoring Methodology
Develop a Site Specific SAI Metric
2) Assign value/weight to each page and site action
Visitor A
Understanding Resistance:
1 pt
Learn About Causes:
1 pt
Watch Demo:
3 pts
Visitor B
Treatment Goals:
2 pts
M O D E L
SAI Score by
Geography
Homepage Bounce:
0 pts
View ISI:
1 pt
Brand Support Enrollment:
2 pts
visit 1
visit 2
visit 1
visit 2
In general, the order in which a site is consumed is less important than
the specific content explored
• The SAI is not impacted by the order of specific pages viewed
• Each of the visit paths below generated an SAI score of 3
Average Site Path?
Home Page (0)
Disease Information Page:
What is Disease? (1)
Patient Experience Feedback Page:
Share Your Story (2)
Tests to Monitor Disease (1)
Alternate Disease Information (1)
Step 3: Score Visits
“Weighted” web visits = total visits
-50 100 150 200 250 300 350 400 450 500
Total Web Visits and Visits*QPA Score, Average per Day, April-September 2011
0 1,000 2,000 3,000 4,000 5,000 6,000 7,000 8,000 9,000 10,000 0 1 2 3 4 5 6 7 8 9 10
# of Web Visits April-Sept 2011 by QPA Score
SAI
SAI
3) Score each visit
– Home page visits with no additional page views or site activity
generate an SAI score of “0” and account for over 1/3 of all site
visits and 20% of all page views
– This explains the low correlation between Rx and web activity –
Steps 4-5: Aggregate Data
Zip-level data proves to be valid in the model
4) Aggregate Scores by Geography
– Since the visitors are generally
anonymous, aggregating at the
SCF (3-digit zip) provides an
acceptable way to correlate with
sales
– 72% of patients fill their
prescriptions within the 3 digit
SCF of their home*
5) Aggregate Sales, Market Events, Other
Promotions by Geography
– We will model SAI vs. sales and
require this at the SCF-level
– The model will control for market
events and other promotional
activity
Step 6: Create Model
Calculate Rx Impact of SAI-Scored Site Visits
6) Quantify Impact on Sales (Rx) Over
Time
•
Calculate appropriate time lag
between activity and Rx in a
longitudinal mixed model
•
Create a regression model to
measure the impact to sales
following the site activity – using
the weighted site visits variable
as the predictor
•
Control for other promotion and
activity, including personal
details, co-pay / voucher
registrations & activations,
market events, and all other
promotion
REGRESSION MODEL
SAI
Score*
Visits
All Other PromoDetails
VouchersRegistrations
Market Events
Exploratory Analysis
SAI Weighted Visits Stronger Correlation to Rx
Brand Rx Brand Rx
Brand Rx Forward 1
SAI * Visits
SAI * Visits
•
Weak relationship between visits
& Rx (correlation <0.5)
•
Stronger relationship between
SAI-scored visits and Rx
(correlation 0.71)
•
Strongest relationship when
accounting for 1-month lag from
visits to Rx outcomes
June Rx
Model Data
•
Level of Analysis: 3-digit zipcode, month
–
3-digit zip code (SCF)
–
Month
–
Brand-R TRx (based on physician office zip) (Month
t
)
–
Web visits Month
t-1
–
SAI Scored visits Month
t-1
–
Physician Details, Samples, Tele-Details (Month
t-1
)
–
Physician Marketing Email, DM, Mobile (Month
t-1
)
–
Patient Marketing Touches including Display impressions, Paid
search clicks (Month
t-1
)
–
Patient Support Program Registrations (Month
t
)
Model Structure and Output
SAS PROCEDURE PROC MIXED
RANDOM
INTERCEPT
MODEL
TRx = SAI Visits, Details, Direct Mail Touches, Email
Sent, Display Impressions, Paid Search Clicks, Co-Pay
Card Program Registrations
Functional Form
Quadratic (sample output below)
Predictor
First-Order
Second-Order
Sample Units
Predicted
TRx
SAI * Visits
0.01166590
(0.00000090)
60
.70
Details
0.13090000
(0.00001450)
230
29.3
Direct Mail
0.02310000
(0.00006320)
90
1.6
Display Impressions
0.00015430 (0.00000000071)
7,500
1.1
Hypothetical Data
Visits
Average SAI
Score Per Visit
“SAI Scored Visits”
Visits*SAI =
TRx Resulting
from SAI
1Value of Initial
Rx due to SAI
21,000
1.0
1,000
10.8
$1,077
1,000
2.0
2,000
19.7
$1,973
500
3.0
1,500
15.5
$1,547
5000
0.25
1,250
13.2
$1,318
Sample Application of the Predictive Equation of Impact of Visits*SAI on TRx
Quantity vs. Quality
Same # of Visits Can Have Different Rx Impact
•
SAI scored visits yield a different impact than visits alone
•
At 1,000 visits, depending on “quality” of the visit, Rx impact ranges
from 10.8-19.7. However, at 5,000 visits of lower content value,
Rx impact is only 13.2
Implication: A search strategy that drives high volume
of lower-SAI traffic may not be as desirable as search
that drives less volume of higher-SAI traffic
Insights
Predictive model shows impact of SAI on Rx
Key Finding
Details
SAI score is highly
predictive of future
BRAND-R TRx volume
–
Higher the SAI score of visits, the more Rx generated
in the following 4 week period
–
Higher SAI scores reflects types of pages viewed
Type of website
activity has varying
impact on future Rx
–
Disease information and payment assistance searches
drive the highest level of overall Rx, including new Rx
–
Re-contact and stay-connected page activity resulted in
growth in TRx, but lower than disease information
search
–
BRAND-R treatment and education page views had the
lowest impact on future Rx
Paid and organic
search drove high
number of visits to
BRAND-R.com
–
Paid search involving the term “cost” or “payment
assistance” generates the highest SAI score
–
Searches that comes through “needymeds.com” or
disease-specific organization sites resulted in higher
SAI scores
• Evaluating SAI by keyword, for example, allows us to predict and optimize
based on projected ROI (not just click through or registration goals)
Keyword Channel Impressions Clicks
Click Through Rate Cost QPA* Visits Avg QPA TRx
Generated Value ROI
chronic disease Google 166,893 1,273 1% $11,100 203 1.66 2.33 $19,040 $ 2 1.7:1
chronic disease MSN 146,614 2,934 2% $7,599 19 1.35 0.22 $1,808 $ 0 0.2:1
chronic disease #2 Google 89,685 320 0% $3,544 110 2.41 1.27 $10,393 $ 3 2.9:1
chronic disease #3 Google 37,939 263 1% $3,088 65 2.40 0.75 $6,163 $ 2 2:1
chronic disease #3 MSN 69,851 403 1% $2,378 26 2.36 0.30 $2,472 $ 1 1:1
abbreviated Google 276,435 1,636 1% $17,022 693 2.04 7.65 $62,504 $ 4 3.7:1
abbreviated MSN 183,452 342 0% $1,906 77 2.02 0.89 $7,294 $ 4 3.8:1
abbreviated disease Google 40,117 310 1% $3,148 56 1.93 0.65 $5,313 $ 2 1.7:1
disease treatment Google 25,362 418 2% $4,962 142 1.78 1.64 $13,383 $ 3 2.7:1
brand chemical name Google 131,441 3,658 3% $14,119 1,482 1.94 15.31 $125,070 $ 9 8.9:1
brand package insert Google 2,680 284 11% $1,266 285 3.03 3.25 $26,560 $ 2 1 21:1
brand Google 67,608 7,979 12% $9,477 6,153 2.54 37.71 $307,990 $ 3 2 32.5:1
brand cost Google 2,597 244 9% $419 197 7.30 2.26 $18,486 $ 4 4 44.1:1
brand with Google 6,894 339 5% $1,235 378 3.89 4.28 $34,968 28.3:1
Actions Taken in Web Visits April-September '11 QPA Scored Visits from Source and ROI Estimate SAI
SAI
SAI
Traditional Web Optimization
Merkle’s Way of Optimization
New Metrics for Search
•
Within paid search, the treatment and disease state related referring sources,
e.g. needymeds.org, webmd.com and righthealth.com had high SAI scores per visit
•
Brand-R should secure additional visits coming from these sites
Application to Search
Focus on driving traffic from higher SAI sites
Paid Organic