Return on Quality Improvements in Search Engine Marketing

Nadia Abou Nabout and Bernd Skiera (Goethe University Frankfurt at the time), Journal of Interactive Marketing 26(3):141–154, August 2012. DOI 10.1016/j.intmar.2011.11.001.

Ingested via the hub research pass against Most wanted #17, the advertiser's side of the ad auction. Read at abstract level only. The article is closed access (ScienceDirect returned 403; Semantic Scholar lists no open PDF), so this page holds the abstract as published by RePEc/IDEAS and the publisher. It does not hold the data description: which advertiser, how many keywords, and what period are in the full text and are not recorded here.

What it asks

Every practitioner source on Quality Score says the same thing: raise quality and you pay less per click. The paper tests that belief on the advertiser's side. In a quality-weighted generalized second-price auction (search-ad-auction, gsp-auction-paper), a quality improvement raises the advertiser's weighted bid. That lowers the price per click only if the ad keeps its rank. If the ad moves up, it now pays the price set by a stronger competitor, and the price per click usually rises.

The authors split the effect of a quality improvement on costs and profits into these parts with a decomposition method, applied to empirical data.

Findings (from the abstract)

Of all quality improvements observed:

• 5% led to a higher price per click for the keyword;
• 100% led to more clicks;
• 53% led to higher total search-marketing costs;
• 37% led to lower profits.

The recommendation follows from the mechanism: advertisers gain if they lower their bids after a quality improvement, so that the gain shows up as a cheaper click at the same rank rather than as a more expensive click at a higher one.

Why it matters here

It is the first source in this spoke that looks at Quality Score from the advertiser's account rather than from the platform's documentation (google-ad-rank) or the auctioneer's revenue (quality-score-auction-estimation). And it cuts against the practitioner claim the spoke has never been able to source: that a better score means a cheaper click. In this data the better score bought more clicks every time, but over a third of the time it lost the advertiser money, because the auction converted quality into position and position costs more.

What it does not give is the number #17 asks for. It counts how often a quality gain raises costs; it does not say how much of a CPC gap between two keywords the score explains. The widely quoted tables (a score of 1 costing about 400% more than a score of 5) are still unsourced. The data are also from before 2012, on a platform whose ranking inputs have changed since.

Caveat on reading. The direction of these findings is clear from the abstract. The sample, the platform and the time window are not, and a reader who needs them must go to the full text.