Online advertising

Hal Varian April 6, 2009


Problems with advertising
‡ You¶ve got to do it ‡ It is unreasonably effective ‡ Nobody knows what really works
± Hard to reach potential customers ± Hard to grab enough of their attention

‡ It¶s difficult to tell how effective marketing is
± Experimentation is costly: ³Does it play in Peoria?´ ± Leads to a portfolio strategy

‡ But technology is changing all of this«advertising is becoming scientific

Broadcast radio
‡ Killer app: ship-to-shore ‡ Broadcast: a nuisance«
± Could that bug become a feature? The Euphoria of 1922«like the Euphoria of 1999 ± What¶s the business model for broadcast radio?
‡ ‡ ‡ ‡ ‡ Government provision (BBC) Private contributions (public radio) Hardware producers subsidize content (RCA) Tax on vacuum tubes (Media tax) Advertising ± the least favorite alternative


Advertising model
‡ First radio commercial
± Apartment rentals in NYC

‡ Challenge to advertising
± Could only survive in areas with dense population ± Need enough potential buyers to pay for content that people would want to listen to ± So only viable in big cities

AT&T to the rescue ‡ Sent radio signals over copper and rebroadcast nationwide ± Built national networks ‡ Didn¶t have to make very much per person to cover cost of producing content ± National brand marketing ‡ TV took this model over when it arrived SIMS .

DVDs. 2 cents for 1 minute. satellite radio: pay per view. and consume what you want SIMS . 20 cents for 10 minutes ± Conclusion: TV network is paid 20 cents an hour of content for your attention ‡ Too cheap. That¶s why trend towards viewer pays for video content ‡ Trend towards cable TV. in general. very loweffectiveness business ‡ Cost measured per thousand impressions (CPM) ± Broadcast TV: $10 CPM for 30 second impr ‡ Superbowl 30 second spot: $25-40 CPM ± Around 1 cent per 30 sec impression.CPMs ± the problem ‡ Advertising is.

Typical CPMs ‡ Outdoor: $1-5 CPM ‡ Cable TV: $5-8 CPM ‡ Radio: $8 CPM ‡ Online ± Display $5-30 CPM ± Contextual: $1-$5 CPM ± Search: $1 to $200 CPM ‡ Network/Local TV: $20 CPM ‡ Magazine: $10-30 CPM ‡ Newspaper: $30-35 CPM ‡ Direct Mail: $250 CPM SIMS .

Online advertising ‡ Search ads (Google Adwords) ± Related to search terms ± Not intrusive. yet highly relevant ‡ Contextual ads (Google AdSense) ± Related to content on page ‡ Display ads (various) ± Standardized ad shapes with images ± Normally not automatically related to content but may be related to audience SIMS .

Pricing for online ads ‡ CPM = cost per 1000 impressions ± Typically used for display ads ± Common denominator for ranking ‡ CPC = cost per click ± Typically used for search + contextual ± Predict prob of click. predict prob of conv. rank by eCPM SIMS . rank by eCPM ‡ CPA = cost per action (or acquisition) ± eCPA = effective cost per action ± Used for contextual.

yet not obtrusive ‡ Has expanded rapidly in last few years ‡ Expectation of further growth« SIMS .Search advertising ‡ Highly effective since it reaches people when they are interested in a topic ‡ Relevant.

Aligning incentives ‡ Google sells ad impressions (views) ‡ Seller generally wants clicks and ultimately conversions (purchases) ± Rank ads by how much advertisers are willing to bid for impressions: ± Bid/click x clicks/impr = bid/impr ± But may want too many impressions ± Answer: disable if eCTR is too low SIMS .

since default cost per click is only ~5 cents SIMS .Pricing personalized ads ‡ Highly differentiated ads: huge pricing problem ± Google used to use impression-based pricing ± Solution: use keyword auction-based system ± Difficulty: depends on competition ‡ Auctions with a lot of competition: prices get pushed to market value ‡ Auctions with little competition can be very valuable.

g..Measurability ‡ Huge advantage in measuring effectiveness ± Pay per click ± Conversion tracking ‡ Value of ranking system ± Ranks ads by bid x ad quality ± Show ads likely to get the most valuable clicks in the most prominent position ± E. ³best ads get best exposure´ ± Creates a virtuous circle: individuals want to click because results are relevant SIMS .

negative keyword ± Follow guidelines for content ‡ Implement conversion tracking if desired SIMS .Ads system from viewpoint of advertiser ‡ Choose your creative text ± Standardized size ± Various optimization tips available ‡ Choose your keywords ± Exact. broad. phrase.

Creating an AdWords Ad 15 .

cont. ‡ Choose your maxCPC ± Value of a click = prob of sale * profit per sale ± Can estimate how many clicks you will get ± Discounter will set actual CPC to minimum amount necessary to retain position ‡ Set daily budget ‡ Let µer rip SIMS .Setting up ad campaign.

exclamation points..g. plurals get more clicks and more conversions than singulars. ³free´ is good. ³Diamonds´ more valuable than ³diamond´ ‡ How much to bid ± Measure cost-per-acquisition and/or ROI SIMS .Challenge to advertisers ‡ Keyword choice ± This is the most critical ± Market efficiencies: high CTR words have high prices ‡ What matters is the cost effectiveness: the ROI ‡ Avoid question marks. can use SEOs ‡ E. repeating product name is good ‡ Choose keywords carefully.

How to bid? Determine value per click ‡Probability of purchase x profit margin ‡Impression value. return visits. Determine relationship between cost and clicks ‡How much do you have to pay to get x clicks? ‡Find profit maximizing point ‡Find bid that gets you to this point ‡ Equivalently use incremental cost per click SIMS . etc.

Profit max bid ‡ Don't want to move up: ± Don't want to move down SIMS .

00 52.80 1.20 0.75 174 124.64 0.78 1.90 0.43 0.80 0.Example Max Est Est Click Est CPC Clicks Cost value Profit CPC ICC 2.00 13.93 1.50 137 84.30 1.06 1.20 0.00 50.20 1.25 119 68.70 0.50 17 3.00 69 32.66 0.62 0.00 198 155.00 0.00 0.47 0.00 50.32 0.20 SIMS .72 1.00 36.00 1.00 43.10 1.00 1.75 38 12.71 1.29 1.34 1.00 25.57 0.

If you can measure performance. why set budget? ‡ I will give you $2 for every $1 you give me ± How many times to you want to do that? ‡ Display advertising is different ± Can catch attention through different media ± Search is generally done on one engine ± Though can catch attention via other means SIMS .How to set budget? ‡ Don't.

Advertiser optimization ‡ Choose good creatives. follow tips ‡ Determine your value per click ± Can use Google¶s conversion tracking ± Or use your own tool ‡ Maximum bid = maximum value per click ± Shade down depending on competition ‡ Minimums ± Bid: ~5 cents ± CTR:~2% SIMS .

Factors affecting revenue for search engine Monetization (RPM) = Revenue Queries Revenue Clicks Revenue Clicks x ( 1K ) = x Clicks Queries Queries w/ Ads Queries Ads Queries w/ Ads Clicks Ads = x x x = CPC x Coverage x Depth x CTR per Ad Price Quantity Quality SIMS .

more ads ‡ Increase CTR ± Show more relevant ads ± Show high quality ads in prominent positions SIMS .Increasing revenue ‡ Increase CPC ± Create higher conversion prob for advertiser ± Capture more value by increasing competition ‡ Increasing coverage ± Get more keywords ± Match more broadly (affects CTR) ‡ Increase depth ± Get more advertisers.

synonyms.Technology ‡ Engine to match ads to queries ± Choose most relevant ads using various criteria (plurals.) ± Returns up to 11 ads to show on a page ± Promotion policy: depends on CTR performance and CPC ‡ Auction engine ± Ranks ads by eCPM. records payment SIMS . etc.

Engineering challenge: Predicting CTR ‡ Dizzying set of factors could affect clickthrough ± Country. targeted text vs query. time of day. « ‡ How does one automatically figure out which factor is more relevant? ± How to update model quickly in face of change ± How do you estimate CTR for not-yet-shown ads? ‡ Huge machine learning model for predicting clicks and other aspects of ad quality 26 .

sliced and diced along many dimensions ± In real time ± Hundreds of millions of rows of data per day ‡ Need to have up to date budget information about hundreds of thousands of advertisers in real time while running many thousands of auctions per second 27 .Statistics collection & reporting ‡ Advertisers want to know how they are doing: ± Serving is inherently distributed and global ± But advertiser view is inherently global ± Want data summarized.

. so little Oracle« ‡ We use MySQL« ‡ and we divide and conquer by partitioning customers onto separate database servers 1 1 2 1 1 . . . .So many advertisers. 1 Clone to scale to capacity same schema and same data n 1 1 Shard to scale to # advertisers same schema but different data 28 .

adjustments made in payment ‡ Very simple to set up SIMS .Contextual Advertising ‡ ³Content ads´ ± Use keyword matching engine to pick ads related to content on page ± Matching algorithm tweaking to deal with multiple content on page ± Auction needs tweaking to deal with position ± Generally lower CTRs. lower conversion performance.

document. } google_show_ad().google_ad_height + ' scrolling=no></ifr' + 'ame>')' + '&url=' + escape(w.googlesyndication.write('<ifr' + 'ame' + ' width=' + w. w.And how we serve the ads function google_show_ad() { var w = window. 30 .google_page_url) + '&hl=' + w.google_ad_url = 'http://pagead2.google_ad_width + ' height=' + w.

AdSense for Content ‡ Contextually-targeted ads ‡ Example: ± What ads would you show? ‡ Buy some cheese ‡ Look for other kinds of cheese ‡ Recipes? Or diet? 31 .

AdSense Challenges ‡ Need to know the context of the page fetch the target URL ‡ Need to understand what it¶s about natural language processing ‡ Need to find the best matching ads how do you even define ³best match´? (a competitor¶s ad would be bad!) ‡ Do this all a gazillion times a minute «serving users on every continent «in over 20 languages «for millions and millions of users 32 .

Display advertising ‡ Prime example: banner ads ± 1000s of advertisers ± 100s of big publishers ± Prices set by negotiation ‡ New kids on the block (mostly for remnant inventory) ± Ad networks ‡ Managed ‡ Make arrangements with ad networks ± Ad exchanges ‡ Unmanaged ‡ Two sided market 33 .

displaying an advertisement. a location on an electronic map such as on Google Maps. or entire webpage. Expanding ad: An ad which changes size and which may alter the contents of the webpage. actual moving video clips are displayed. except that instead of a static or animated image. Trick banner: A banner ad that looks like a dialog box with buttons. Pop-up: A new window which opens in front of the current one. and appearing in or over. Pop-under: Similar to a Pop-Up except that the window is loaded or sent behind the current window so that the user does not see it until they close one or more active windows. Polite ad: A method by which a large ad will be downloaded in smaller pieces to minimize the disruption of the content being viewed Wallpaper ad: An ad which changes the background of the page being viewed. Video ad: similar to a banner ad. Mobile ad: an SMS text or multi-media message sent to a cell phone. SIMS ‡ ‡ ‡ .Display formats ‡ ‡ ‡ ‡ ‡ ‡ ‡ Floating ad: An ad which moves across the user's screen or floats above the content. It simulates an error message or an alert. Map ad: text or graphics linked from.

Display ad pricing ‡ Generally on a CPM basis ± Nominally there is a rate card ± In practice there is negotiation with a variety of terms ± Ad exchanges and networks use auction like mechanism ‡ Number of players ± 100s of web sites ± 1000s of advertisers ‡ Serving is done by 3rd parties ± Publisher side ± Advertiser side ± Economies of scope: serve for your site. serve for others SIMS .

or content ± Behavioral targeting ± Contextual targeting ‡ ‡ ‡ Tuning and optimization based on results. Reporting impressions. clicks and interaction metrics ± Reconciling ad side and pub side Prices for basic serving tend to be a few percent of ad value ± Action is in value added services and remnant ads SIMS .Ad server functionality ‡ ‡ Uploading ads Managing ads according to differing business rules (based on contract) ± Frequency capping ± Ad rotation and sequencing ± Collision avoidance ‡ Targeting ads to different users.

Amazon collaborative filtering Surround Session: Enable a single advertiser to own ad space for a user's entire session Frequency Capping: Limiting the number of times a user views any one advertisement SIMS .g. ± Example: readers of financial site see ads for Mercedes on financial site and extended elsewhere ‡ Retargeting: Recognizing consumers who have already seen or expressed interest in a product or brand and showing targeted advertisements in different media or locations ± Search Retargeting: Showing search-relevant ads on a page visited immediately after a search execution ‡ ‡ ‡ Audience Amplification: Finding similar users to those you're already targeting ± E.Behavioral targeting ‡ Audience Extension: Extending valuable inventory by labeling an audience in one environment and recognizing that audience elsewhere..

Type of data ‡ Demographic: zip code. gender (ZAG) ‡ Contextual ‡ Transactional ‡ Geographic ‡ Psychographic ‡ Affinity SIMS . age.

Interest based advertising SIMS .

Syndication network ‡ Can syndicate all types of ads ‡ Search ± AOL = both ± Ask Jeeves = ads only ± Yahoo (formerly) = search only ± MSN was served by Overture/Yahoo ‡ Negotiated revenue share with partners ‡ Can do this for all 3 types of ads SIMS .

SIMS . Can get immediate feedback on test campaigns.Reverse channel ‡ Use Web to measure effectiveness of local TV/radio/newspaper ± Look at URL queries ± Automobile companies ± Allows for much more precise measurement of effectiveness ‡ Cosmetics companies: search term research for product design ‡ Auto companies: use web queries to measure media/campaign effectiveness.

make better decisions SIMS .Wall Street and Madison Ave ‡ Finance: came of age in 1970s with rise of quants ± Psychology -> rationality/quants -> quant psychology ‡ Next step: is marketing quants ± Psychology -> rationality/quants -> quant psychology ‡ Pricing keywords. measuring effectiveness-> quantify marketing.

Sign up to vote on this title
UsefulNot useful