Loading ...« Locally Focused Web Site Tips for Connecting With Your Customers: Part One | Home | Weekly Wrap Up - 11 April 2008 »
Geo-Targeting Primer - Part 3 - Systems
By Cameron Ferroni | April 10, 2008
As the last post in my Geo-Targeting series I wanted to bring together the user and advertiser perspective and highlight how that impacts the design of ad serving platforms. The short version is - you’ve seen what the advertiser wants and what the user wants - now the trick is translating all of that into a usable software product that delivers on the promise!
It starts with the ad management system - this is the piece that the advertiser interacts with. From here an advertiser can create their ads, determine where their ads are displayed, choose how much to spend for their ads, and track all of their ROI. Geo-targeting plays a large role in all of these steps. Starting with the creation of the ad - a world class ad system should allow the advertiser to have different ads for different regions, it should allow for the inclusion of geographical information directly in the ad - either dynamically putting the name of the city/state/neighborhood depending on the user query, or hard coding it if that is what the advertiser wants. When it comes to where the ad should be displayed - the advertiser should be able at a minimum to choose any of national, state, DMA, or city where their ad should run. A truly world class product should evolve beyond that however, and allow targeting by neighborhood, by lat/long, and by arbitrary shaped regions within a geography. Yes, try to wrap your head around that UI challenge, and the complexities it would bring, but at the same time, you know we need it. When it comes to budgeting, over time we are going to see a world where geo-targeting has a direct impact on how much people are willing to pay for ads, and we have to make this possible - a highly targeted ad for a restaurant in a specific neighborhood when the site knows I’m searching for restaurants in that neighborhood should cost more (and be worth more) than a generic restaurant ad for a city when I’m searching in a given neighborhood. And obviously, once you have given all of this flexibility, the reporting engine must be capable of slicing and dicing impressions and clicks based on the underlying geographical data - I as an advertiser need to know if my city based ad is driving good enough ROI, or if I should be spending more on my highly targeted ads because they provide the best leads. Most of this is actually pretty straightforward in software - with one huge caveat - doing this well, in a way that is easy enough for advertisers to use, and provides them with the data they need, while still providing the right level of flexibility/customizability - that is the trick.
Once all of the ads are into the system, then the biggest challenge arises - namely - how does the system deterministically ensure that the right ads are shown at the right time. When it comes to geographical relevance, that is where the magic really happens. It starts by figuring out where the user is. This is the most straightforward (albeit not foolproof). Massive databases of IP addresses exist where software can easily guess at the location of the user based on their address. As we’ve discussed this is only accurate about 80% of the time, but it is a start. Second, you have to try to factor in user intent - did the user specify some geographic keyword - and if they did - could you tell? I might type in wall street restaurant - does this mean I’m looking for a restaurant in New York’s financial district, or maybe I’m looking for a restaurant on Wall in Seattle? Maybe I’m on newyorkrestaurants.com, well, then that is definitely a big indicator. As you can quickly see, deducing all of this just by looking at IP + Site + Content + Query is a fun problem. Assuming for a second that you’ve devised a set of heuristics that can translate into a best guess for user intent, now you have to layer in the advertiser desire - when and how do they want their ads to show up - have they picked a lat/long radius that falls within the user intent map? Have they picked the right neighborhood? What if they are just on the border?
OK, now you’ve determined a set of ads that fall within the user intent. The last step is actually presenting these to the user - now you have to decide how to order them - do you order them based on bid price? Or based on relevancy? How do you determine relevancy? Is the closest business to the user intent map shown first? What if it is for a French Restaurant, but the user was looking for Italian? This relevancy algorithm is something we will talk more about in an upcoming post, for now suffice it to say that it is the key to all three constituents - a good relevancy algorithm gives users the best results, provides advertisers with the best leads, and provides the publisher with the best monetary gains.
So as you can see - this is a great example of where a task that seems trivial to a person - figure out the best ad to show someone - is actually quite challenging for a computer. But that’s why we have smart people that actually program the computers!
In my ongoing Local initiatives I took my wife to a park yesterday that she had never been to, so that counts, and I went to a new restaurant in the International District - Gourmet Noodle Bowl - great hot pot and potstickers - we’ll be going back for lunch to try more noodles!
Topics: Neighborhood, Maps, Local Advertisers, Local Search, Advertising, PPClick, Content |