Early Tests Show Flaws in Facebook Zip Code Targeting

San Francisco-- Facebook's debut of Zip code targeting last week was met with general applause, but early analysis suggests it can't be trusted for accuracy. At least not yet.

During a session at SES San Francisco this week, XA.net CEO Rob Leathern presented data from Facebook ad testing system optim.al that hinted at accuracy problems for Zip code-based ad targeting. Specifically, Facebook's estimated addressable audience in several Zip codes in Fremont, CA and Pleasanton, CA did not match to census data for those places.

In the case of at least one Zip code (94566), Facebook's user estimates are grossly inflated compared to the U.S. Census count. Facebook pegs its audience at 808,000 for that Pleasanton postal zone, whereas census data puts it at 41,000.

In an email exchange after the panel, Leathern noted the problem is not exclusive to Facebook and said he is optimistic about the eventual accuracy of geotargeting on the site.

"Zip code targeting based on IP has historically been less than 50 percent accurate in display advertising, and Facebook is not much different (yet)," he said. "Facebook does have the advantage of a lot of user-supplied location data as well, and we're pretty confident from talking to them that they're in the process of figuring it out and improving the targeting capabilities for advertisers."

He added the failure is not Facebook's directly, but that of the geographic and IP data that inform its estimates.

Leathern was not the only speaker at this event to suggest flaws in Facebook's Zip code targeting. "Zip code targeting doesn't look like it works that well," said Marty Weintraub, founder of aimClear, an online marketing agency.

Before the introduction of Zip code targeting late last week, advertisers could target geographically only to cities, states, and provinces. They can now deliver individual ads to as many as 2,500 Zip codes.