Upload
john-hathaway
View
219
Download
0
Embed Size (px)
Citation preview
Patterns in Online Shopping Behavior(and possible links to call centers)
Peter S. FaderProfessor of Marketing
The Wharton School, University of [email protected]
www.petefader.com
The basic business model for e-commerce is no different than for many traditional goods & services
1. Generate/understand store traffic
2. Convert visits into purchases
3. Have customers buy repeatedly over time
4. Have them spend more at each transaction
1. Generate/understand traffic
• Changes in visit patterns over time (Moe and Fader 2003)– How does behavior evolve as the visitor gains experience with the
site?– What can visiting patterns tell us about purchasing propensities?
• Modeling browsing behavior at multiple websites (Park and Fader 2003)– What can we learn about our customer behavior at one site from
observing their behavior at other sites?
Data description
• Focus of the traffic and conversion studies:– Media Metrix’s PC Meter panel of 20,000 web users– Two sites (Amazon.com and CDNOW, from 3/98 – 10/98)– Aggregated to household level and daily level
• Define purchases as any visits with “confirm-order” in URL
HHID DATE TIME ACTIVE DOMAIN URL
10145 3/17/98 9:47:02 PM 2 www.amazon.com /exec/obidos/isbn=0060972777/kcweba/0728/0973004/501562:10145 3/17/98 9:49:36 PM 48 www.amazon.com /exec/obidos/isbn=0859652424/kcweba/0728/0973004/501562
10145 3/18/98 5:33:59 PM 120 www.amazon.com /exec/obidos/quicksearch/query/nscp0873199410145 3/18/98 5:35:59 PM 19 www.amazon.com /exec/obidos/isbn=0671024868/7024/1486849/08491310145 3/18/98 5:36:18 PM 8 www.amazon.com /exec/obidos/shopping/basket/7024/1486849/08491310145 3/18/98 5:36:26 PM 9 www.amazon.com /exec/obidos/subst/home/home.html/7024/1486849/084913?:10145 3/18/98 5:47:02 PM 57 www.amazon.com /exec/obidos/order/form/page2/7024/1486849/08491310145 3/18/98 5:47:59 PM 30 www.amazon.com /exec/obidos/confirm/order/7024/1486849/084913
10145 3/20/98 8:55:35 PM 26 www.amazon.com /exec/obidos/subst/home/home.html/7024/1486849/08491310145 3/20/98 8:56:01 PM 21 www.amazon.com/exec/obidos/recommendations/past/purchases/7024/1486849/084913:
1. Generate/understand traffic
• Changes in visit patterns over time (Moe and Fader 2003)– How does behavior evolve as the visitor gains experience with the
site?– What can visiting patterns tell us about purchasing propensities?
Frequency of site visits
0%
10%20%
30%
40%
50%60%
70%
1 2 3 4 5 6 7 8 9 10+
Number of Visits
% o
f H
ouse
hol
ds Bookstore
CD Store
Visit dynamics: aggregate pattern
• Visits/visitor ratio suggests that consumers are visiting the site more frequently over time
BOOKSTORE CD STORE
Months 1-4 Months 5-8 Months 1-4 Months 5-8
Total Numberof Visits
5402 5899 1729 1890
Number ofUnique Visitors
2693 2717 988 920
Visits / Visitor 2.01 2.17 1.75 2.05
How has visiting behavior evolved at Amazon?
Mean evolution is close to 1.0 (E[cij] = 0.998)
• All shoppers have an initial visit rate (
• After each visit, the rate is bumped up or down by an “updating multiplier,” (c)
0%
10%
20%
30%
40%
50%
60%
0.02 0.06 0.1 0.14 0.18 0.22 0.26 0.3 0.34 0.38
Rate of Visit ()
% o
f Hou
seho
lds
r = 0.324 = 16.857
Distribution of Initial Visiting Rates
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.05 0.25 0.45 0.65 0.85 1.05 1.25 1.45 1.65 1.85
Updating Multiplier (c)
f(c)
s = 2.299 = 2.304
Distribution of Updating Multiplier
slowing down
speedingup
Forecasting validation
0
2000
4000
6000
8000
10000
1 4 7 10 13 16 19 22 25 28 31 34
Week
Rep
eat
Vis
its
Actual Static Model Evolving Visit Model
Visit frequency and purchasing propensity
• Mall shopping research: there is a relationship between visit frequency and purchasing propensity (Celsi and Olson 1988, Janiszewski 1998, Jarboe and McDaniel 1987, Roy 1994)
Bookstore Conversion Rates
Conversion Rate Decreasing Frequency Increasing Frequency
Infrequent Visitors 11.1% 10.9% 11.3%
Frequent Visitors 16.6% 14.6% 18.6%
CD Store Conversion Rates
Conversion Rate Decreasing Frequency Increasing Frequency
Infrequent Visitors 4.8% 3.8% 5.7%
Frequent Visitors 5.6% 4.0% 7.6%
1. Generate/understand traffic
• Changes in visit patterns over time (Moe and Fader 2003)– How does behavior evolve as the visitor gains experience with the store site?– What can visiting patterns tell us about purchasing propensities?
• Modeling browsing behavior at multiple websites (Park and Fader 2003)– What can we learn about our customer behavior at one site from observing
their behavior at other sites?
Modeling browsing behavior at multiple websites
• What is the nature of the associations among visit patterns across sites?
• How much does combining information improve our view of future visit behavior?
Site B?
time
Site A?
time
Associations across browsing patterns
Site Btime
Site Btime
Site AtimePerson 1
Site AtimePerson 2
• Need to account for
– Co-incidence (similarity in arrival times)
– Overall rate propensities (similarity in latent rates)
Model results
• Fit models over the first 4 months of data
• Forecast the number of previous non-visitors who first visit during the second 4 months
Actual vs. Predicted % of Previous Non-Visitors
0
5
10
15
20
25
30
35
Amazon Barnes & Noble
% o
f P
revi
ou
s N
on
-Vis
itors
Actual Proposed Model Independent Model(s)
The basic business model for e-commerce is no different than for many traditional goods & services
1. Generate/understand store traffic
2. Convert visits into purchases
3. Have customers buy repeatedly over time
4. Have them spend more at each transaction
2. Convert visits into purchases
• Across visits (Moe and Fader 2003)
– How do conversion rates change from visit to visit?
– Role of store visits: browsing, searching, or directed buying
• Within a visit (work in progress)
– What will the shopper do after the current page? Buy? Exit?
– Response to “interstitial” promotions
Conversion rates across visits
P P
t1 t2 t3 t4 t5 t6
?
P P
t1 t2 t3 t4
?
P P
t1 t2 t3 t4 t5 t6
?P
Conceptual model
What contributes to the Net Effect of Visits (Vij)?
• Directed-Buyer: consistently large baseline effects on purchasing
• Searcher: variable effects of visits that accumulate
• Browser: variable visit effects that do not accumulate
)( Threshold
Purchasing
)( PurchaseLast Since
Visits ofEffect Net
)( PurchaseLast Since
Visits ofEffect Net
ijij
ijij
TV
Vp
What is the probability that person i will buy at visit j (pij)?
Ranking likely buyers
• Rank customers based on four months of observed data using different models
• Observe actual purchasing behavior in next visit
Top 10% according to… Actual Conversion Rate
Historical conversion rates 29.3%
Beta binomial model 33.0%
Logistic regression 33.0%
Logistic regression (2 seg) 28.0%
Proposed conversion model 37.0%
Conversion rates within a given visit
• What drives the probability of purchase?
– Baseline for this visit (from V-P model)
– Page-to-page effects:
> Types of page being viewed
> Duration of page views
> Number of different products and categories viewed
• What drives the probability of exit?
The basic business model for e-commerce is no different than for many traditional goods & services
1. Generate/understand store traffic
2. Convert visits into purchases
3. Have customers buy repeatedly over time
4. Have them spend more at each transaction
3. Have customers buy repeatedly over time
Observations about transaction dynamics (Fader, Hardie, and Huang 2003):
– Individual-level repeat buying behavior appears to be quite random at the start, but settles down towards a “steady state” over time
– So it’s hard to gauge someone’s long-term repeat buying tendencies from their early behavior at a given site
– But the aggregate pattern of “steady state” behaviors can be described relatively early and very accurately
It is possible to make strong statements about the lifetime value of a given group of customers, but individual-level descriptions are not very trustworthy.
The basic business model for e-commerce is no different than for traditional goods & services
1. Generate store traffic
2. Convert visits into purchases
3. Have customers buy repeatedly over time
4. Have them spend more at each transaction
4. Have them spend more at each transaction
• It is very important to separate out transactions from purchase quantities
• CDNOW forecasting model (Fader and Hardie 2001)
– Number of albums purchased per transaction is governed by a “coin-flipping” process
– Everyone has a unique “coin,” (i.e., tendency to buy multiple albums)
– This process does not appear to vary over time among repeat buyers: almost all dynamics are due to transaction effects
Purchase quantity models are straightforward and reliable when transaction dynamics are handled separately.
Moving from online shopping to call centers…
• Do you know when your customer(s) will contact you next?
• Do you know what they will be trying to accomplish?
• Are you prepared to handle their requests?
• Do you have a sense of what’s going to add (or extract) value during their next contact?