Incremental Clicks Impact Of Search Advertising

SearchEngineLand has a very interesting post on how Google researchers have attempted to measure the economic tradeoff of organic listings and paid search advertisements.  The real jewel of the article is the link to this paper (PDF) however.

That paper gives great insight into how Google is calculating what Hal Varian terms VPC (value per click).  The very last equation of the paper informs us that advertisers halt their ad campaigns unless:

(v – c) / v > r*(1-IAC)

where IAC is estimated in the paper, by Bayesian methods (Gibbs Sampling and Slice Sampling), to be about 0.89 (mean expectation).  The LHS isn’t exactly innocent:  rewriting it as 1 – c/v, it is a function of the fraction of marginal revenue Google gets to keep.  That is, if a click has marginal value to an advertiser of v and Google gets to keep marginal CPC c, then c/v is Google’s (percentage) take.

A few worked examples will make the point: we can re-arrange c/v < 1 – r*(1-IAC) ['Google's take will be no more than the RHS, and with proper tuning can be that amount without causing an advertiser to pause a campaign, behaviourally]. ‘r‘ is the relative (conversion) value of an organic click to a paid search click.  Since IAC is *estimated* in the article, we can eliminate it from the equation by substitution, and for any value of r, learn Google’s percent take.  For example, if r = 1, then Google keeps about 90%.  If r is only 0.1 however, Google could keep 99%.  When r is as much as 10, they keep nothing at all.

Thus, Google’s interest in mining its logs to estimate IAC is quite understandable!  They operate a bazaar, and charge rent on the booths.  They can count the number of *transactions* that occur in their mart (up to uncertainty about how many clicks result in actual conversions), but they cannot see the amount of money that changes hands.  If they could learn what their customers are making (marginal VPC), they could price the stalls in the mart so that they receive an optimal fraction of the money changing hands there.  Hence, the extreme inetrest in researching the two variables that determine that — r, the marginal rate of substitution of an organic click’s value for a paid ad’s one, and 1 – IAC, the ‘cannibalism factor.’

Of course, telling your advertisers that IAC is high (and 1 – IAC 10%-ish, though quite skewed to higher values) helps retain their business.  That, however, is not Google’s only motive here.  VPC may be unknowable — but what is revealed here is that Google thinks the (maximal) CPC / VPC ratio is, in fact, only determined by two factors: r and IAC.

They have estimated the latter — IAC — and revealed its median value is very high (98%) and because of skew its expected value quite a bit less (90%).  Naturally, the marginal rate of substitution will vary from advertiser to advertiser.  I would guess most clicks leading to an advertiser’s site have similar value, though organic clicks may be worth less, since they will contain information queries that may not ‘convert’ as well as someone clicking on an ad.

In any event, Google has told us about their two key revenue drivers, and how they relate to optimising their business.  I suppose the next order of business, then, is to understand what factors affect the one they *haven’t* told us about quantitatively.  Clearly, the relative importance of organic listings to paid search will depend on the relevance and quality of those listings.  The interesting point we see here, is how Google conceives of the tradeoff of revenue and those relevance factors — namely, in terms of marginal rate of substitution between competing pathways to a website obtaining a click.