The generalised second price auction, described in a somewhat mathematical paper by University of Stanford economists, is the dominant auction mechanism for paid search. It is often misunderstood. I’m pretty certain that there’s features of it that I don’t properly understand, too. However, understanding the auction is important for effective bidding and also explains a lot about the value of paid search companies. Think about the deeper mechanics of the auction, because, along with the number of advertisers involved in a market, the details of the auction will affect revenue. In other words - if Yahoo! uses Google adverts, they could get more or less from the market than Google does, because of the way that they handle bids and bidding; the auction rules and mechanics affect revenue.
Material Disclosure: I am not, and have not been, employed by any search engine companies. My 30 year old degree is in Cybernetics Science, and my main academic research involved dynamic thermal models of houses and measuring low energy housing performance. Anything that I’ve got right about economics is through The Economist, New Scientist, and a few influential books - see the bibliography in the next article in this series. I’m pretty happy with this analysis - it affects how I perceive the search engines, but YMMV. This kind of publishing is the closest that I get to a peer review of my thinking. Well, that and paid search results…
Simple Auctions
One of the most basic auction mechanisms is that every participant offers a bid and the highest bidder wins. Another favourite, sometimes used for government contract bidding, is when all participants offer a sealed bid, and the second lowest bid is the winner. The rules of the auction determine who is the winner - and those rules therefore influence strategy. In the first example, you’d need to bid the most to win; that strategy would lose you the winning position in the second type of auction. This coarse example should make it clear that understanding the model that the auction uses will affect bidding strategy.
The original Yahoo!Search Marketing auction, when it was GoTo.com and Overture and prior to the rollout of the still fairly new Panama system, used a slight variation of the simple “highest bidder wins” auction. Bids were ranked and the advertising slot you got was based on the rank of your bid. For example, if you bid the second highest amount, then you showed in slot 2. You could consider it an auction for each slot, where the topmost bidder wins each time, and is then removed from the auction for lower places.
This model, or variants of it, tends to be used in the second and third tier paid search vehicles. It is simple, easy to understand and has several crucial flaws:
- Simple auctions don’t maximise revenue (bid x impression volume per position x advert CTR, is not optimised)
- It distracts advertisers from focusing on relevance in favour of technical bid twiddling.
- Search Engines can’t broaden the keyword, which reduces revenue
There’s one more twist to a standard auction that is common in paid search. The offer that “you pay only one penny more per click than the advertiser below you is bidding.” If you are number one with a bid of $10.00, and number 2 is offering $0.23, then you pay $0.24 per click, not $10.00 - at least, not until advertiser two notices what they can do to your advertising budget with a simple bid change… This “penny more per click” offer also has an interesting effect on auctions and revenue.
Revenue Optimisation And Simple Auctions
Let’s illustrate an example. We’ll pretend to look at a generic keyword like “holiday”. The number one advertiser is a new and little known business. They get a CTR of 2% because they have focused on flogging the tagline for the business and have failed to mention the word “holiday” in the advert. Instead their message is something about brighter futures and feeling less bloated; totally incomprehensible as a holiday advert. The number 2 advertiser has an 8% CTR. Their advert mentions “holiday” and they have a well known brand name. If the order of the two adverts was swapped, even though the second advertiser is bidding less, then the publisher would make more money - because searchers will further increase the click rate for that more relevant second advert. Even if the previously number 2 advertiser pays no more per click than they were paying in position 2, to gain position 1, the higher CTR advert in this position will yield a higher revenue for the publisher. This is most easily understood as thinking that the CTR for the advert will increase when moved to position 1. More money makes for happier shareholders and bigger bonuses.
There is an additional incentive to consider a different auction method, for the search engine. It even helps advertisers, too. Strictly, it helps aware advertisers. If you have any clue what messaging means, and can divine the searchers’ intent, then you can build a better advert than your less aware competitors. If you can deliver adverts that attract more clicks per page, then you have arguably improved the search engine experience for users. That should make the search engine into a more popular property because users can know that adverts as well as organic results are pretty well optimised.
Now, it isn’t always true that higher CTR means higher revenue for the advertiser - I’ll address that elsewhere. It’s a subtle point, but quite important for certain types of business.
It’s the Relevance, Dummy
The old Overture bidding model spawned a collection of companies offering automated bidding tools. With 96 chances to update the bid per day, you could work out what bidding patterns your competitors used and do things like targeting large gaps, and bidding high under competitors, so that their adverts cost more, and so on. The emphasis was not on servicing the search users, but on technological twiddling.
I’d typically find, on taking over an Overture account, that the adverts were all very bland, very generic, with low CTR - but someone had invested a lot of time in bid management. Completely the wrong focus for both advertisers and publishers.
Open Auctions
Some auctions are open - all the participants can see what the currently winning bid is. This is typical of animal, car and property auctions. When there is only one item to be won, you only need to know the topmost bid to know the winner. However, paid search advertising is subtly different. Instead of wanting a single winner, the goal of the auction is to find multiple advertisers and rank them by some criteria. The simplest extension of that is to take the “winner takes all” auction and extend it. If there are three advertising slots, then the top three bidders take the slots, in the order of the bids.
However, that model has a weakness. It assumes that the bid and the value to the advert publisher are the same thing. They aren’t, in many practical cases. To illustrate this, look at the numbers below:
Bid: 1.00
CTR: 1%
Revenue: 1.00 per 100 impressions or a CPM of 10.00
Bid: 0.5
CTR: 2.5%
Revenue: 1.25 per 100 impressions or a CPM of 12.50
The second advert, although offering a lower bid, has a higher CTR and is therefore a better value to the publisher. It is also, interestingly, more likely to be considered valuable by users of search advertising - more of them click the second advert than the first, suggesting that it has a higher perceived relevance.
By making the auction open, you can make some assumptions about competitor adverts and bid higher. However, rationality often flees at this point. For example, brand owners may insist to their agency or staff that they appear first in results. If this strategy can be identified then competitors can drive up the cost to the advertiser at little expense to themselves. Most people won’t click on competitor adverts and they are even less likely to do so if the advert is deliberately made unattractive.
This example shows how a business can drive up advertising costs of a competitor at no significant cost to themselves. An open auction is not what you could call a fair system. It leaves advertisers feeling dissatisfaction. On the other hand, this effect is advantageous to the publisher, because the CTR on the first advertiser is likely to be reasonably high, and their bid has been forced up. Brand terms aren’t the only searches that yield conversions, and they arguably cannibalise organic search - choosing to use or not use a simple auction just because of the effect on Brand keywords would be perverse. There are better reasons to consider changing to another model - as evidenced by none of the top three continuing to use a simple, open, auction.
Ignore The Man Behind The Curtain
AdWords says that it is a Pay Per Click advertising system. But is it?
You certainly appear to pay with every click in the keyword advertising campaigns. If one click costs $0.10, then two clicks costs $0.20 (or so). The value that you pay is quantised at the penny interval for each click. So it sure looks as though you are buying clicks.
Let me give you a disturbing vision… Google is actually operating a CPM bidding system. Yup. Your bid position is a consequence of the value you give to Google, per thousand impressions - but quantised to a penny per click.
By ranking adverts according to the CPM, Google can overcome many of the defects of simpler bidding systems:
- Search users get (mostly) more relevant adverts
- Revenue is optimised
- Keywords can be matched to a wide range of search queries
There are some consequences that drop out and further subtleties in choice of the auction mechanics that affect the revenue that can be extracted.
In the next article, I’ll look at what happens with a CPM model and quantised CPC payments, what the “Quality Score” is likely to mean for auctions, and various other implications such as the assumed CTR before you have a statistically relevant volume of impressions.
Updates
2008-07-29 - The Register Describes AdWords AutoMatch - partially pre-empting the next part of this series. Minor edits for clarity.
