Facebook’s Miscalculated Metrics
Article referenced in post: http://www.wsj.com/articles/facebook-says-it-found-more-miscalculated-metrics-1479303984
On November 16th, Facebook reported that it discovered more flaws in its calculation methods related to advertising. Facebook conducted an internal audit of its metrics analysis and found that the metric failed to properly estimated weekly and monthly reach of advertisers’ posts. This is very troubling to advertisers because they pay for exposure and if Facebook cannot properly estimate its reach, it cannot properly charge advertisers. This incident, along with earlier revelations suggesting that Facebook’s metrics were flawed, has led to public demands for third party verification. There have even been calls for Media Rating Council, an independent watchdog organization, to audit Facebook. Facebook tracks over 200 metrics and uses them to create pricing methods to charge advertisers. Facebook’s audit found that some metrics undercounted while others overcounted. This is not just a problem for advertisers; it inhibits Facebook’s ability to properly price its advertisement slots. For a long time now, Facebook has been regarded as a leader in advertising due to its large base of users. This has led many businesses to buy ad space from Facebook. Significant challenges still exist between Facebook and advertisers; many of these issues revolve around how much information Facebook makes available to advertisers and the public.
In Chapter 15, search markets were covered. Sponsored search markets have many pricing schemes like paying-per-click and setting prices through an auction system. Concepts like clickthrough rate were introduced. Clickthrough rate refers to the number of clicks per hour an advertisement will receive. The metrics Facebook uses most likely attempt to measure and accurately estimate these factors. The Facebook algorithm that measures the clickthrough rate is most likely very complex, but the fundamental ideas regarding market clearing for advertisers, Vickrey-Clarke-Groves mechanism (VCG), and Generalized Second Price auction (GSP) are probably relevant to the algorithm’s computing. Another topic discussed was the underlying assumption that the clickthrough rate is fixed; in practice, it most likely is not. One phenomenon related to a fixed rate that is the likelihood that users look at the thumbnail and choose not to click on it. Since Facebook and other companies get paid per click, as opposed to per impression, they will not gain revenue from impressions. This will have implications for bids for slots in auction systems. It is unclear if Facebook’s algorithm and pricing schemes account for this. It most likely makes some assessment of impressions because, as mentioned before, Facebook tracks over 200 metrics. Also, The complexity of Facebook ads fundamentally varies from the complexity of search engine advertisements discussed in Chapter 15. Facebook is a social media platform which does not accept queries in the way that Google and other search engines would. Google uses queries to match the ad slot with advertisers who wish to reach people searching for particular queries. Facebook must rely on an individual’s social media page likes, interactions, posts, etc. to determine which advertisers will be most interested in what ad slots.