The Economics of Internet Search Slides adopted from a talk by: Hal R. Varian Google’s Chief Economist UC Berkeley Economics Prof. Sept 31, 2007

Search engine Advertisement Search engines are highly profitable Revenue comes from selling ads related to queries 99% of Google’s revenue come from ads Yahoo, MSN also uses similar model Banner ads (Doubleclick) Standardized ad shapes with images Loosely related to content Context linked ads (Google AdSense) Related to content on page Search linked ads (Google Adwords) Related to search terms Banner ad Content ads

Search engine Advertisement Search engines are highly profitable Revenue comes from selling ads related to queries 99% of Google’s revenue come from ads Yahoo, MSN also uses similar model Banner ads (Doubleclick) Standardized ad shapes with images Loosely related to content Context linked ads (Google AdSense) Related to content on page Search linked ads (Google Adwords) Related to search terms Top-of-organic ads RHS ads Organic search results 3

Search engine ads Ads are highly effective due to high relevance But even so, advertising still requires scale 2% of ads (“impressions”) might get clicks 2% of clicks might convert into sales So only 4 / 1000 who see an ad actually buy Cost Per Million impressions (CPM) will not be large But this performance is good compared to conventional advertising! Search technology exhibits increasing returns to scale High fixed costs for infrastructure, low marginal costs for serving CPM: actually, Cost per Million impressions. Web: $2 PPI Conventional advertising: $10 CPM (superbowl far more expensive) Bcast model: Radio was initially to communicate with ships; not a good pt2pt so turned into entertainment. Lots of debates on how to support it. Eventually commercially support radio was the winning model.

Advertising industry economies Entry costs (at a profitable scale) are large due to fixed costs Very low incremental cost for 2nd ad User switching costs are low 56% of search engine users use more than one (but 44% use only one…) Advertisers follow the eyeballs Place ads wherever there are sufficient users, no exclusivity Hence market’s structure is likely to be A few large search engines in each language/country group Highly contestable market for users No demand-side network effects that drive towards a single supplier so multiple players can co-exist No exclusivity: 90% of users use Google AND 80% use Yahoo AND 75% use MSN… Ads can play multiple roles: An impression could have javascript that goes to doubleclick $5 to set up an AdSense account at Google!

What services do S.E.’s provide? Google is yenta (matchmaker) Matches up those seeking info to those having info Matches up buyers with sellers Relevant disciplines Computer Science: Theory of information retrieval Economics: The assignment problem: Match efficiently providers to consumers Assignment: match efficiently providers and consumers

The advent of the web By mid-1990s IR algorithms were very mature Very little difference in IR competitions (mostly CIA sponsored) The Web comes along IR researchers were slow to react CS researchers were quick to react Crucial NSF grants: 1993 to UIC - NCSA (Mosaic); 1994 to Stanford (Digital Libraries Initiative) Link structure of Web became new explanatory variable PageRank improved relevance of search results dramatically Mature: CIA sponsored competitions of fast retrieval; winner was only slightly ahead of the next place

Google’s birth Brin and Page tried to sell algorithm to Yahoo for just $1 million Yahoo did not buy it; they believed that search was in a commodity business, nothing new to find while Yahoo is in the Directory business, it adds value to information Most people thought this way, banner ads were the only (boring) game They formed Google in late 1998 with no real idea of how they would make money Put a lot of effort into improving algorithms Started selling intranet services (they still do) Then tried selling keywords (like everyone else) via negotiation

Why online business are different Online businesses (Amazon, eBay, Google…) can continually experiment Hard to really do continuously for offline companies (they sell a product) Manufacturing Services Very easy to do continually for online companies (they sell online service) Leads to very rapid (and subtle) improvement Learning-by-doing leads to significant competitive advantage kaizen (Jap.)= “continuous improvement” Every moment are 400 experiments going on at Google trying new ways of promoting ads etc.

Business model (aka: how to make money out of it?) Initially: sell keywords All SE’s and Dir’s did it; it does not scale Then: Sell search ranking But people did not like their ranks be ad-driven They split the results: some organic, some ad-driven (2000) GoTo Ad Auction GoTo’s model was to auction search results Changed name to Overture, auctioned ads Google liked ad auction, tried to improve on Overture’s model (2001) Lawsuit by Overture (2002) settled in 2003 Google licenses technology from Yahoo (bought Overture for $1.63B) Initially they tried selling the engine to search client’s intranet Then they tried to sell keywords - price negotiable b/w google and seller - IT DOES NOT SCALE Overture founded by CalTech student who took a course from economist Charlie Plott (auction specialist), who thought of auctioning off search results – went off to Bill Gross (idealab) But people did not want their results to be auctioned off. So they decided to split off the results! Some organic, some auctioned ads.

Original Overture model Rank ads by bids Ads assigned to slots depending on bids Highest bidders get better (higher up) slots High bidders pay what they bid (1st price auction) (about $0.50) How-to bid example: Conversion Rate = 1% sales per click Net profit per sale = $20 If you bid PPC=$0.20/click, you are expected to break even 11

Google ads Ads are selected based on query+keywords Ordering (Ranking) of ads based on expected revenue Negotiation-based (used to) Google saw Overture’s model in the Fall 2001. Came up with variation above. Initially: Top 2 ads sold by negotiation (price per impression) Right 8 ads sold by auction; TODAY: All go through auction, as we will see from the analysis later Auction-based

Google’s auction version Rank ads by expected revenue This makes sense: revenue = price * quantity Expected revenue = bid * expected clicks But you sell impressions, so price /impression = Price /click * clicks /impression Each bidder pays price determined by bidder below him Price = minimum bid necessary to retain position Motivated by engineering, not economics People would do it by trial & error anyway… Overture (now owned by Yahoo) Adopted 2nd price model in 2007 Increased their revenues There is another strategy VCG pricing: charge each advertiser with the cost he imposes on the other advertisers: What would happen if you took each bidder out of the system? In this case the optimal value would be to bid your true value per click, no matter what the others are doing.

Google and Game theory When everyone bids optimally, we have Nash equilibrium Basic principle of equilibrium: each bidder prefers the position he/she is in to any other position Gives set of inequalities that can be analyzed to describe equilibrium Inequalities can also be inverted to give values as a function of bids The highest value people end up in higher positions Gives set of inequalities that can be analyzed to describe equilibrium I.e., I get more value in, say, pos. 3 than I get in pos. 2. If I increase my bid by 10%, I will increase my sales, but I pay more. Are the extra costs worth the extra value I pay?

Implications of analysis Basic theoretical result: incremental cost per click has to be increasing in the click through rate. Why? If incremental cost per click ever decreased, then someone bought expensive clicks and passed up cheap ones. Similar to classic competitive pricing in economics Price = marginal cost Marginal cost has to be increasing

Simple example Suppose all advertisers have same value v for click Two cases: Undersold auctions: There are more slots on page than bidders. r = The minimum price per click (the “reserved price”) (say, 5 cents). Oversold auctions: There are more bidders than slots on page. Last bidder pays price determined by 1st excluded bidder. There are 8 ad positions on the RHS. Say there are only 4 advertisers: undersold. The last person pays 0.05$ per clicks. The guy above him has to pay enough to capture the extra clicks. Say they are 20 advertisers: oversold

Undersold pages Bidder in each slot must be indifferent to being in last slot v = value you get ps = price you pay at slot s xs = number of clicks in slot s r = reserve value xm = number of clicks in last position Payment for slot s = payment for last position + value of incremental clicks

Example of undersold case Two slots x1 = 100 clicks x2 = 80 clicks v = 0.50 r = 0.05 Solve equation P1 * 100 = 0.05 * 80 + 0.50 * (100-80) p1 = 14 cents, p2=5 cents Google’s Revenue = 0.14 * 100 + 0.05 * 80 = $18

Oversold pages Each bidder has to be indifferent between having his slot and not being shown: So For previous 2-slot example, when 3 bidders: ps=50 cents for each slot, and Google’s revenue = 0.50 x (100+80) = $90 Revenue takes big jump when advertisers have to compete for slots!

How Google determines number of ads shown Showing more ads Pushes revenue up, particularly moving from undersold to oversold But relevancy goes down (irrelevant ads will sneak in) => Users click less in future (“ad blindness”) Optimal choice Depends on balancing short run profit against long run goals Top-of-organic vs RHS ads Initially top ads sold negotiated, but RHS revenue much higher Now all are auctioned, but for top ads are only selected if they are very relevant If you only see RHS ads: they were good enough to be shown, but not so relevant to be on top of organic Ad blindness: the average American sees 3000 ads a day; they ignore most of them! TODAY: top ads are shown because they also have relevance, (as measured by click through, too) not just are winners of auctions Avoid ad blindness that way!

Other forms of online ads Contextual ads AdSense puts relevant text ads next to content Treats the content of the page as a query! Publisher (content owner) puts some Javascript on page and shares in revenue from ad clicks Display ads (aka “Rich Media Ad”) Advertiser negotiates with publisher for CPM (price) and impressions (at specific time, for specific audience, for up to X impressions/viewer) Ad server (e.g. Doubleclick) serves up ads to pub server Are more sophisticated than simple ads (e.g. max num of impr/IP addr.) Ad effectiveness Increase reach Target frequency Privacy issues AdSense gives incentives to publishers to produce new content! (even if Google were to make no money off these ads) Google tries to buy DoubleClick; claims it is complementary business Yahoo is power into display ad, google in contextual ads.

Monitoring and fine-tuning

Opportunities for employment “Marketing as the new finance” Availability of real time data allows for fine tuning, constant improvement Market prices reflect value Quantitative methods are very valuable We are just at the beginning… To increase revenue, look at your “creatives” (the text around the ad) Diamonds (want to buy some) is a more expensive keyword than Diamond (school report) Repeating the query in the ad is better than not Google will let you use 3 creatives, and will use the one that gives more traffic (on the fly experimentation) Landing page should be related to the ad they clicked on – many advertisers miss it! Once on the landing page, make it easy for them to buy it. Maybe it may not be worth being the top bidder. Maybe too expensive. Wine market – words to advertise: Cabernet, etc. Every wine advertiser does that, it is expensive But “gourmet gift basket” gets a lot of relevant clicks!!! Less obvious are cheap Keep it simple, keep it clean But still try to figure out how to monetize YouTube: should not be intrusive Froogle is revampled, it was not working well. Product Search now Behavioral search does not seem that relevant!! If you are shopping, you are behaving different than your usual search. Lots of research on Behavioral Search, not much in terms of development