Search, Search Mechanisms and Price Discrimination in Electronic Markets

1
Search, Search Mechanisms and Price
Discrimination in Electronic Markets

• Ravi Aron

2
Search

• In the beginning
• There were no brands or so the economists would
  have us believe.
• If two stores that sold exactly the same products
  were located next to each other what would the
  average consumer do?
• If she found a product (at one of the stores)
  that was priced at a level that exceeded her
  willingness to pay, would she try the other
  store?
• What if the product was a well known brand?
• Why bother to Brand products at all?

3
Quiz Are you a good Microsoft?

• City A
• There are two malls right next to each other
  owned by two different entities.
• City B
• The two malls are separated by a distance of 50
  miles.
• One of the two malls in City A offers a high
  price shoe retailer a deal wherein he would pay a
  huge premium in exchange for the Mall owner not
  hosting any other shoe retailer. The other offers
  no such deal.
• Should the retailer accept this deal?
• What if one of the malls in City B were to offer
  this deal?

4
Search

• This problem attracted the attention of a famous
  economist, Hotelling in 1929.
• He solved the problem and presented his paper
  before of congregation of Mathematicians (the
  American math. Association). He concluded that if
  two merchants were to sell exactly the same good,
  they were better off by locating right in the
  middle of main street cheek by jowl next to each
  other.
• Fortunately, he was wrong.
• In an unrelated coincidence the stock market
  crashed a couple of months later.

5
Product Differentiation

• Just as firms may seek to put some distance
  between themselves in geographical space, they
  seek to differentiate themselves in Product Space
  too.
• The idea of product differentiation as way of
  preserving (monopoly) pricing power caught on in
  the early 30s.
• This is simply not true. The practice of
  differentiation goes back at least to early Rome,
  when Caesar expressed a preference for fish
  caught on the East Bank of Tiber (even the Romans
  seem to have had qualms about the West bank
  even in those days).
• The idea being that two cars were never alike
  even though they were both black and made by
  Ford. And so you would not compare prices because
  you were comparing Apples and Oranges (or PG
  with Unilever).

6
What Search Costs Can Do

• Subsequently economists have established that
• Sellers can adopt monopolistic pricing in the
  presence of even small search costs.
• Search costs can contribute significantly to
  buyer acceptance of sub-optimal choices and high
  seller profits.
• Search costs support seller collusion.
• The Business medias drumbeat when Web commerce
  took off was
• Prices will fall to Marginal Cost
• The inefficient producer will be forced out of
  the market
• Buyers will indulge in price comparison over the
  net every time they shop
• The commoditization of products has begun
• Search Engines shall inherit the e-world

7
Price Comparison

• Prices have not fallen to MC.
• Price comparison does not seem to be the driving
  factor behind e-buying.
• Amazon Vs. the Shopping Bots (game, set and
  match to Amazon).
• Brands continue to have power and sellers seem to
  have pricing power although there has
  definitely been a decline in (mean) prices in
  several product categories.
• How does search impact market outcomes

8
(No Transcript)
9
(No Transcript)
10
Search Costs, Sellers and Buyers

• Other things being equal, buyers will compare
  prices between two sellers and choose the cheaper
  one.
• If the cost of searching were to be more than the
  gain that accrues to the buyer because of the
  search costs, the buyer will not search.
• Sellers therefore, will attempt to do one of the
  two things
• Make sure that other things are not equal
• Make sure that search is costly

11
Search and the Seller

• Sellers as a group stand to loose considerably in
  introducing electronic search mechanisms that
  lower search costs.
• The seller who is first to market with a search
  mechanism stands to gain considerably (Ex
  Sabre).
• As a result search mechanisms are often sponsored
  by sellers ex. Travelocity, Buy.com etc.

12
Restricting Buyer Search

• Sellers often pioneer search mechanisms. However
  they try to control buyers search behavior.
• Emphasize product information rather than price
  information.
• Prevent buyers from finding out precise prices
  for well defined product features (SNAP.Com)
• Make it difficult for buyers to simultaneously
  compare both price and product features
  (Travelocity, Expedia).
• Levy fees for higher levels of usage.

13
PriceScan

• PS was launched in April 1997 at an initial
  investment of 16,000.
• Allows buyers to search for products and do price
  comparison based on
• Product features
• Brand
• Make, Model and Manufacturer
• Sellers initially balked at having their prices
  displayed.
• PS took the old world technology route of
  hunting up published catalogs and displaying
  prices.
• Sellers retaliated by denying access to their
  sites a move that hurt them badly since PS was
  attracting 9,000 hits a day (up from 1000 a day
  in 6 months) and is growing at 23 a month.

14
The Info Structure of Search

• Sellers realized that simply refusing to have
  your prices displayed will not work.
• How will sellers protect their pricing power?
• By observing that search is not just about price
  comparison.
• In addition to prices, buyers compare product
  features.
• More importantly buyers also compare product
  quality and seller reputations for service
  reliability etc.
• The four dimensions of search, price, product
  features, quality and seller reputations
  constitute the Info Structure of Search.

15
Four Dimensional Comparison

• While it may be difficult to prevent direct
  comparison along any two dimensions, it is always
  possible to prevent effective simultaneous
  comparisons along all four dimensions.
• What makes four dimensional search difficult?
• Inadequate quality comparison standards
• Assessing seller reputation for reliability
  requires user feedback
• Human ability to process information is limited
• Search engines employ automated data gathering
  and indexing methods and cannot process human
  feedback and abstract a quality rating from
  descriptive text generated by humans

16
Models of Search

• Search Engines initially used methods borrowed
  from Information Retrieval.
• They simply went out and scanned the universe
  for all available descriptions of products and
  then indexed them using automated means.
• This produced varying levels of Recall and
  Precision.
• In Information retrieval on the web there were
  two predominant models initially.

17
Precision and Recall

• Precision Suppose you were looking for
  information on Zoom lenses. If 100 documents were
  returned by the search engine and 50 were indeed
  about zoom lenses then the precision of the
  search engine was said to be 50/100 0.5.
• Recall if out of a total of 500 documents in
  the universe about zoom lenses, your SEs
  returned 50 relevant documents then the recall is
  said to be 50/500 0.10 0r 10.

18
Extent of Human Intervention

• Search engines also differed on the means used
  for indexing documents, based on the extent of
  human intervention involved.
• At the one extreme you have Yahoo which is a hand
  crafted directory. Yahoo covers a very small
  fraction of the universe but claims to offer
  great accuracy in returning stuff that is
  relevant to the user.
• AltaVista on the other hand, covers a great many
  documents, but can often bring back documents
  with a poor degree of fit.
• Yahoo trades off (higher) Precision for (lower)
  Recall while AltaVista takes the opposite
  approach.

19
Search Techniques

• There are new and improved versions of these
  engines and techniques now being employed by
  search engines.
• The Citation model of search Retrieve documents
  and when some documents are repeatedly chosen by
  users, allot higher weights to these so as to
  make them more representative of the content of
  the domain of search.
• If some documents have numerous pointers from
  very many other documents, then consider those
  documents as important sources of information.
• ex Inktomi, Northern Lights, Google

20
Search Techniques

• The expert model of search
• Identify some users that are experts within
  certain domains (say one of you is an expert on
  Industrial Auctions).
• When those users make choices, allocate higher
  weights to the documents that they choose.
• The next time someone queries a domain for the
  same information, give them the documents based
  on the new weights. Learn more from their
  behavior and modify the weights.
• ex Excite_at_home, Lycos etc.
• A refinement of this model is used by Meta Search
  Engines Ask Jeeves, HotBot.com etc.

21
When is Pricing Power most Vulnerable?

• When the dimensionality of search is reduced by
  characteristics of the product.
• Commoditification When a product cannot be
  easily differentiated
• Few or no product features
• the search collapses to a single feature
  (quality variation is signaled by price)
• Well defined standards exist that accurately
  specify seller reputations (B2B industrial goods)
• Ex Industrial input goods, steel, industrial
  glass, chemicals etc.
• This is where B2B procurement sites are most
  effective

22
Pricing Power Vulnerability

• Intermediaries that act as Info Mediaries.
• When third party entities can provide quality and
  reliability ratings that decrease the
  dimensionality of search, Pricing power is under
  threat.
• ex. CNet.com seller reliability ratings
• Amazon / eBay Seller reputation and reliability
  ratings
• Amazons editorial and buyer reviews provide both
  quality and extent of fit information/ratings
  between buyer needs and product attributes.

23
Between a Bot and a Hard Place

• Shop Bots are Intelligent Agents or simply
  automated mechanisms that scan merchants sites
  in real-time to generate price quotes.
• Merchants can use these for the same purpose as
  buyers scan the products and prices of
  competitors.
• Several merchants selectively try to deny access
  to these mechanisms by permitting buyer agents
  to scan their products even as they deny seller
  agents access.
• This has inspired Weblets that claim to offer
  secret counter-counteragents that can slip past
  the gates, sometimes by lying about their origins

24
Dynamic Pricing

• Ichoose A software firm in Dallas that is
  developing ever-shrewder software in order to
• enable online merchants scan rival sites
• adjust prices on the fly and
• steal customers, all without leaving a clue
• Liaison Technology in Austin, dispatches sneaky
  scouts that mimic human usage to get past a
  site's sentries.
• Sellers fear bots that can understand customer
  needs and make suggestions on the fly i.e.
  dynamic pricing

25
(No Transcript)
26
Bots and Info Structure of Search

• How do Shopping Bots handle the multiple
  dimensionality of search problem?
• Answer by Specializing in product categories.
• Here are few examples
• CompareNet Recently purchased by Microsoft,
  CompareNet is an effective search engine for
  office hardware, furniture and supplies.
• Jango/Excite Product Finder Has a feature called
  Find Reviews which helps buyer assess product
  quality.
• MySimon MySimon uses a slightly different model
  of search. Using Virtual Learning Agent
  technology, it scours a thousand merchants for
  everything from computer hardware to gourmet
  chocolate and updates its ratings based on
  aggregate measures of fit with buyers needs.
• BottomDollar A price-Bot, it finds hot deals on
  a wide range of products.

27
Bots and the Info Structure of Search

• DealPilot The just-in-time Bot. The companys
  DealPilot Express service, a browser plug-in,
  will let users check competing prices -- complete
  with shipping, taxes and any duties -- at the
  click of a button.
• Frictionless Commerce Inc. and Active Research
  Inc. Now, a new breed of recommendation engine
  and services from companies that don't just look
  at price. They use a computational technique
  based on generating aggregate statistical fitness
  scores for determining the product components
  consumers are looking for, factoring in concepts
  such as quality and service. In short, the
  services add value to the recommendation
  equation.

28
Revenue Models for Search Engines

• Pay to be listed Some search engines ask sellers
  to pay a fee to be listed on the S.E.s site.
• ex. Yahoo!'s shopping software will initially
  sample prices only at vendors that have paid the
  Web site to be listed as part of Yahoo!'s
  "on-line mall."
• The bargain-hungry buyer can later click on a
  special button to get access to a broader price
  list.
• Advertising Revenue Widely adopted model.
  PriceScan.com, MySimon.com, Esmarts.com.
• Referral Fee Many sites charge (or attempt to
  charge) a referral fee for sending a customer
  to a merchants site.
• Esmarts - 5-to-8 cut of the sales
• Washington Post Co. - 5 to 15.
• Junglee now Acquired by Amazon unknown
  fraction

29
A Catch 22 Situation?

• Consider a search engines problem If it charges
  a merchant 8-10 for referrals, then for sure the
  merchant has priced at least 8-10 above his MC.
• Those merchants that make profit margins of under
  10 will not list themselves on this engine.
• If there exist one or more search engines that do
  not charge Merchants for referrals, then the
  real low priced sellers will be able to list
  themselves on these engines.
• So, if a SE is profitable, it is not accurate and
  if it is accurate it is garnering sub-optimal
  revenues.

30
Differentiating the Search Offering

• How will the better search engine(s) overcome
  this difficulty?
• Answer the same way that the merchants fight
  price comparison. The commission-based search
  engines will provide superior info structure
  services. Which in turn means that SEs will
  partition the market on the basis of product
  complexity.
• Indeed they have done so. PS, BottomDollar.com
  etc. tend to specialize mostly on the price
  attribute and derive their revenues from
  advertisements.
• MySimon, Lycos.com etc. tend to provide a
  superior match and collect commision revenue.

31
Intelligent Agents Price Discrimination

• The advent of the Shopping Bot seems to make life
  difficult for the seller and a lot easier for the
  buyer.
• Not quite. Intelligent Agents can work for
  sellers too.
• Just buyers want to compare seller prices,
  sellers would love to know how much a buyer will
  pay for a particular product or the buyers
  reservation price.

32
A Question of Pricing

• Suppose I knew that I had exactly 2 buyers for a
  new PDA card reader that I was planning to sell
  on the web.
• Further, I know that one of them values it at
  100 and the other at 90, but do not know which
  buyer values it more (and therefore which one
  values it less).
• Assume that my (unit) cost of producing the
  product is less than 50.
• What is the price at which I should list my
  product?
• Answer ___________
• If you are not sure, have you considered starting
  a Dot Com company? (-

33
The Pricing Problem

• Suppose that there are a 1000 buyers for the PDA
  card reader that I sell on the web.
• Further, I know that each buyer values the card
  reader at an amount that falls between 80 and
  100 (all values are equally likely).
• My (unit) cost of producing the product is less
  than 50.
• What is the price at which I should list my
  product?
• Answer ___________
• What if they all valued it between 40 and 100
  (all values being equally likely) ?

34
Price Discrimination
35
Price Discrimination on the Web

• Let us revisit the pricing problem
• Suppose there are two kinds of buyers that exist
  for my product.
• Price Sensitive Buyers who value the product
  between 80 and 90 and
• The high valuation buyer who values the product
  between 90 and 100.
• If I could separate the two, I would offer them
  two different price schedules.

36
One Product Two Prices

• Suppose I offer those buyers who are referred to
  my site by BottomDollar.com or PriceScan.com a
  price of ________
• And those that come from the Palm or Mindspring
  site a small (almost costless) freebie (say a CD
  ROM back up of software) along with the same
  product and price it at ______________
• How does the Info Structure of Search help me do
  it?
• What signal does the PriceScan buyer send? Why?

37
Making Inferences about Buyers Based on Buyer
Characteristics

• Site Demographics
• Buyer Demographics
• Collaborative Filtering
• Observed Real-time behavior
• Some techniques such as Data mining are in the
  nascent stages now, which help in establishing
  similarity metrics between buyers.

38
Making Inferences about Buyers
39
How Much Will You Pay?
40
Mapping the Buyer

• Site Demographics Makes the assumption that the
  characteristics of a site also represent the
  buyers tastes.
• The buyer that arrives from a site such as say,
  the Forbes Cool Portfolio collection or Paul
  Fredrcikss custom shop is more likely to pay
  premium dollar for a product.
• Easy to implement but tends to be of limited
  value.

41
Buyer Demographics

• Buyer Demographics Based on the tell-me-
  more-about-yourself approach. Buyers voluntarily
  part with information in exchange for some
  (usually useless) privilege. Merchants use this
  to infer the prices buyers will pay.
• ex 1-800-flowers.com, AOLs e-mall etc.
• Easy to implement, but suffers from poor
  inaccuracy and often, adverse selection problem.
  The naïve buyer will tell you his life story
  while the more savvy one will let you think that
  hes on a shoe string budget.

42
Collaborative Filtering

• Collaborative Filtering Attempts to establish
  the degree of similarity between buyers.
• Sometimes called Recommender Systems
• Linda seems to have the same tastes as Mary who
  bought the Handbook of Industrial Abrasives so
  recommend it to Linda.
• Instead Linda buys a Danielle Steele, the
  collaborative filtering system makes some minor
  modifications.
• ex Amazon.com, Yahoo.com, AOL Market helper

43
Observed Human Real-Time Behavior

• Observed Human Real-Time Behavior The holy
  grail of the Price Discriminators.
• Observe the behavior of a buyer, make inferences
  about her product preferences and proclivity to
  pay and then attempt to modify her behavior.
• Techniques such as Neural Nets to mimic human
  behavior and aggregate statistical models to
  guess where a buyer falls within a known
  distribution of buyers are being tried.
• Yahoo Vs. AOL differing results.

44
Product Customization as a Window into the
Buyers Soul

• The idea is to get buyers to customize their
  products.
• As they make choices about their ideal product,
  the merchant makes inferences about the buyers
  valuation of the product(s).
• Try to fit the buyer in a distribution of buyers
  or establish a metric of similarities between
  buyers.
• Based on how much other buyers paid in the past,
  set a price for the buyer whose indicated a
  certain level of customization.

45
How Much Will You Pay?