How Valuable Are Shopbots?: Panos M. Markopoulos Jeffrey O. Kephart
How Valuable Are Shopbots?: Panos M. Markopoulos Jeffrey O. Kephart
How Valuable Are Shopbots?: Panos M. Markopoulos Jeffrey O. Kephart
Panos M. Markopoulos
Jeffrey O. Kephart
markopou@unagi.cis.upenn.edu
kephart@us.ibm.com
ABSTRACT
The price information that shopbots provide to buyers is
clearly valuable, as it enables them to make a better informed choice of product and vendor. We quantify the value
of this price information to the buyer in terms of the price
dispersion and the buyers brand preferences, and consider
scenarios in which the buyer pays a seller, a shopbot, or some
other third party for price information. As an illustration,
we compute the value of price information of well known retailers in online book markets, using data on price dispersion
and brand preferences reported by Smith and Brynjolfsson,
finding that information about a books price can be about
6% to 10% as valuable as the book itself.
Keywords
shopbots, brand, price dispersion, information value
1.
INTRODUCTION
Shopbotscomparison-shopping web sites that collate information on products from multiple vendorscan be a very
valuable tool for buyers [1, 8]. Typically, they permit buyers to sort product and vendor information along desired
dimensions, such as price, delivery time, or vendor reputation. The most sophisticated shopbots even provide personalized rankings that take into account an individual buyers
product and vendor preferences.
There are two components to the typical business model
employed by shopbots. Like most Internet information services, shopbots typically sell advertising space on their website. In addition, many shopbots make a commission on
sales that result from clickthrough purchases. The main
beneficiaries of the servicethe buyerspay nothing at all.
This is understandable. Today, it would be infeasible to
charge buyers for this service, given the lack of a suitably
widespread micropayment scheme, coupled with the inconvenience to buyers of being forced to deliberate over whether
to pay small amounts of money to get access to product information.
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. To copy otherwise, to
republish, to post on servers or to redistribute to lists, requires prior specific
permission and/or a fee.
AAMAS02, July 15-19, 2002, Bologna, Italy.
Copyright 2002 ACM 1-58113-480-0/02/0007 ...$5.00.
However, in the future, we envision that buyers will employ economically-motivated agents that will purchase physical goods on behalf of human owners, and that in order to
carry out this task they may purchase information services,
such as the product information service offered by shopbots.
In such a future, it will be feasible for the shopbot or the sellers themselves to charge buyer agents for price information.
In order for a buyer agent to know how much it can spend
on product information, it must know the value of that information. Shopbots could use such information to govern
how long to wait for product information from sellers, or the
order in which sellers are contacted. Finally, sellers of price
information (presumably sellers of the physical product, or
the shopbot) could use knowledge of the value of product information to buyers in order to establish a fair price. Thus it
is interesting and important to quantify the value of product
information.
In this paper, our objective is to quantify the value of
product information. We do so by means of a simple model
that captures two of the most common and important dimensions of concern to buyers: price and brand.
Price information is valuable to the extent that price dispersion exists in a market. Despite claims that the Internet is frictionless, significant price dispersion has been
observed in online markets [4, 7, 3]. Furthermore, brand
appears to have a larger influence over purchase decisions
than other variables, according to Smith and Brynjolfsson
[5], who found that even buyers that use shopbots select the
cheapest vendor just half the time.
A similar model that took into account only price information was presented in [10]. That paper mostly considered
gaming issues and showed that product sellers have incentives to sell their price information and that their information price will not be driven down to zero in competition
settings. Our paper tries to quantify the value of product information regardless of which market entity (sellers
or intermediary) takes advantage of it. By adding brand
considerations in our model we hope to better model the
human-shopbot interaction.
In section 2, we introduce a simple model of valuations,
taking brand effects into account, and we derive an expression for the value of price information in terms of price dispersion and brand preferences. In section 3, we calibrate
our model using data on online bookstore price levels and
buyer preferences, as reported by Smith and Brynjolfsson[4].
Next, in section 4, we quantify the value of price information
to buyers under two different assumptions about how much
price information is already known to the buyers. Then, in
2.
OUR MODEL
bi bm +pm
ui um =
Retailer
KingBooks:
A1Books:
Borders:
1Bookstreet:
BN:
Amazon
Shopbot
Market Share
18.6%
19.0%
20.7%
11.2%
14.1%
16.3%
bi bm +pm
bi bm +pm
Fi (pi )dpi ,
bi bm +pm
ui um =
Fi (x)dx
(1)
3.
INTERNET BOOKSTORES
Retailers at a Shopbot
Proportion of
Lowest Prices
29.7%
24.6%
19.3%
16.3%
6.1%
3.9%
(2)
0.2
Rank
Naive
1
2
3
4
5
6
KingBooks: 29.7%
A1Books: 24.6%
Borders: 19.3%
1Bookstreet: 16.3%
BN: 6.1%
Amazon 3.9%
Kingbooks
A1Books
Borders
1Bookstreet
BN
Amazon
0.15
0.1
0.05
$7.5
$10
$12.5
$15
$17.5
Predicted by
our model
Borders: 24.8%
Kingbooks: 22.0%
A1Books: 17.9%
Amazon: 15.5%
1Bookstreet: 11.5%
BN: 8.3%
Actual
Borders: 20.7%
A1Books: 19.0%
Kingbooks: 18.6%
Amazon: 16.3%
BN: 14.1%
1Bookstreet: 11.2%
$20
14.0%
0.2
Borders
Kingbooks
A1Books
Amazon
1Bookstreet
BN
0.15
0.1
0.05
$7.5
$10
$12.5
$15
$17.5
$20
12.0%
10.0%
8.0%
predicted by our
model
naive
6.0%
4.0%
2.0%
0.0%
Borders
KingBooks A1Books
Amazon 1Bookstreet
BN
Figure 3: Reality check: comparing the absolute errors of our model and the naive approach, that says
that buyers buy at shopbots only based on price.
as shown in table 2 and figure 3, but still a significant improvement over the naive assumption that shopbot users
prefer the cheapest product, which, according to [5] is true
only half the times. Differences from the actual data exist due to effects that are not captured by our model, such
as different buyer sensitivity to different price components2 ,
repeated sales effects, special promotions etc.
4.
Z
u(i|j) =
fj (x)
0
Fi (z)dz dx
(3)
$0.3
$1.80
Price Information
Value
$0.25
$1.60
$0.2
$1.40
$0.15
$1.20
$0.1
$1.00
$0.05
$0.80
$0.60
$0.40
BN
1Bookstreet
Amazon
A1Books
KingBooks
Borders
$0.20
ow
un
BN
tre
et
kn
known j
1Bo
oks
azo
Am
Bo
oks
A1
Bo
oks
Kin
g
Bo
rde
r
$0.00
(4)
min
dFS
(x)
u(i|S) =
0
Z
fSmin (x)
Fi (z)dz dx
(5)
6
8
Additional Sellers
Borders
KingBooks & others
A1Books
Amazon
1Bookstreet
BN
10
5.
Value per
Product Sold
$1.2
10%
Borders
KingBooks & others
A1Books
Amazon
1Bookstreet
BN
$1
$0.8
$0.6
8%
Information Value
as a Percentage of
Product Revenue
KingBooks & others
Borders
A1Books
1Bookstreet
Amazon
BN
6%
4%
$0.4
2%
$0.2
Additional Sellers
Additional Sellers
6
8
10
10 %
Information Value
as a Percentage of
Product Revenue
8%
6%
A1Books
1Bookstreet
Amazon
BN
4%
2%
Additional Sellers
2
10
10
would be able to charge the marginal value of their information to the potential buyer. However, this is not always
possible. Figures 6 and 7 refer to the maximum revenue that
an infoseller could possibly extract from the buyer. This is
of course an upper bound that probably cannot be realized.
The most important reason is that the buyer will usually
be able to access the sellers price information through other
means. We introduce an alternative search cost c that the
buyer can incur and learn seller is product price. This cost
represents the time and effort of visiting the sellers web site
directly and the added inconvenience of not having this information as part of the shopbots comparison shopping environment that facilitates the buyers decision making process. This cost can range from a few cents to infinity, in the
case that a seller refuses to provide any price information,
even at his website. The infoseller cannot charge the buyer
more than c, as in that case the buyer would choose to incur
the cost c and learn is price. From equation 5, the average
maximum price that the infoseller can hope to charge the
buyer for price information is:
Z
Z
Fi (z)dz dx (6)
0
Price Information
Value
$0.3
Borders
8%
KingBooks
$0.25
$0.15
$0.1
1BookStreet
Amazon
BN
6%
A1Books
$0.2
KingBooks
Borders
A1Books
Information
Value as a
Percentage of
Product
Revenue
Amazon
4%
1Bookstreet
2%
Alternative
Search Cost
BN
$0.05
$0.2
$0.4
$0.6
Alternative
Search Cost
$1 $1.2
$0.8
Figure 9: Average maximum value of the price information of the six bookstores, assuming that the
shopbot displays for free the information of the
other five, as a function of the buyers alternative
search cost c
Borders
KingBooks
A1Books
Amazon
1Bookstreet
BN
$0.8
$0.6
$0.4
$0.2
Alternative
Search Cost
$0.2
$0.4
$0.6
$0.8
$1
$1.2
$0.2
$0.4
$0.6
$0.8
can extract from a buyer, per product that a seller sells and
as a percentage of a sellers product revenue, respectively.
We observe in figure 11 that unbranded, cheap retailers gain a comparative advantage in a market for product
information: their information is relatively valuable, while
their product revenues are relatively low, as a consequence
of their preference towards price competition. The value
ratios vary starting from 6% for BN.com to 10% for KingBooks3 , showing that sellers that compete mainly based on
price would have greater incentives to participate in such a
market for information. We also see that the value ratios are
high for Borders which even though it enjoys brand power,
maintains low overall price levels.
Finally, it should be noted that the values calculated in
this section assume that the infoseller is allowed to charge
different information prices for different books that a bookstore carries. Pricing is done so that it is always marginally
optimal for the buyer to purchase the price quote. If this
is not possible and a fixed price for the price quotes of a
bookstore is necessary, the expected revenues from price information deteriorate as can be seen in figure 12 for the case
of Amazon. This happens because, now, it is not certain
3
$1.2
gi|S (q) =
Figure 10: Value generated by the price information
of a bookstore per book sold, as a function of the
alternative search cost c. Alternatively, the value
that the price information of a book generates to
shopbot users, before the actual book is sold.
$1
Fi (x)dx
0
1
1
Let gi|S
be the inverse of gi|S , in other words, gi|S
maps a
price quote price to a product price. If q is the minimum
product price among the S sellers, an infoseller that charges
x for seller is price information would have an expected
revenue:
1
Ri (x) = x P r(gi|S
(x) < q),
1
1
(x) is less
(x) < q) is the probability that gi|S
where P r(gi|S
than q. This is simply:
1
Ri (x) = x 1 FSmin gi|S
(x)
(7)
and the first order conditions give the optimal fixed price
quote price:
0
Ri (x) = 0
(8)
6.
$0.25
Fixed price
$0.20
Average price
and expected
profits per buyer
with variable price
$0.15
$0.10
Expected profits
per buyer with
fixed price
$0.05
$0
Additional Sellers
10
7.
CONCLUSIONS
We envision a future where economically-motivated software agents will be employed by humans to simplify their
purchases of physical goods. In such a future it will be feasible for the product seller or an intermediary (shopbot) to
charge buyer agents for price information. In this paper
we were able to quantify the value of product information
by means of a simple model that captures two important
product dimensions, price and brand, using data from the
Brynjolfsson and Smith reports.
We have separated the buyers and the information sellers
view of the problem and demonstrated that due to the multiplicative effect of information, product information value
can be as high as 10% of the product cost itself, in a market
comprised of real online bookstores.
Furthermore, we have discussed the possibility of the emergence of product information markets which would provide
a new business model for the shopbots operation. Product
information markets would contribute to higher information
transparency as sellers would have the incentives to provide
8.
ACKNOWLEDGEMENTS
9.
REFERENCES