As Sir Martin Sorrell said, location targeting in mobile is the holy grail of marketing. Reaching consumers when they are out and about, on the go, interacting with both the digital and real worlds together can actually fulfill the longstanding goal of ‘right ad, right person, right place, right time’.
Location is a big deal. For the history of digital advertising we’ve basically been targeting on a few things, like content, cookie data and search. The big new data set mobile brings to the table is location – a data set that can be as important at inferring intent, demographics, audience segments and other attributes as any we have seen to date. And it brings genuine value to end users – helping them find products and services where and when they want them – a key attribute of the most valuable advertising mediums.
There is a lot of energy and excitement around location targeting in mobile – from agencies and brands that see the enormous potential and from technology companies that are building the next-generation platforms for harvesting and leveraging this data for marketing and advertising.
But there’s one very significant problem that is holding mobile location targeting back – the market has been flooded with bad location data. And many are turning a blind eye to this fact because to acknowledge it would be contrary to their business interests.
There are many sources for actionable mobile location data. The device can share GPS level data (provided the user has opted in to share location with an app or web site). This GPS data is typically represented by latitude and longitude coordinates (lat/long) and is generally very accurate. The network can share data derived from cell tower triangulation which is also quite accurate. There is also user-supplied location data (e.g. when a user provides his zip code when registering for a site or service). Another method of deriving location is IP address analysis, which can range from very accurate to completely inaccurate. An IP lookup can resolve to a terrestrial wifi network (like an airport or coffee shop), most of which are well indexed to location. But an IP lookup can also resolve to a carrier IP address and carrier IPs vary widely in accuracy. Many carrier IP locations are accurate to the metro or zip level, but many resolve to “backhaul” addresses – which are the locations where the carrier aggregates mobile data for transport over fiber trunks in the telecommunication infrastructure — which are generally inaccurate.
The crème of the crop in location data targeting is device-level GPS lat/long data. Highly targeted mobile advertising campaigns that are focused on targeting people in an area the size of a city block or a shopping center or a big box store require this precise lat/long data (and potentially wifi data). The problem is that lat/long data is very scarce. Most industry experts consistently estimate that about 5-10% of mobile ad impressions have lat/long from users who have opted in to share location with an app or site. This makes sense. Most apps and many of the largest mobile consumer apps – like internet radio or game apps — aren’t allowed by Apple to ask for location. One of the rules for app store approval is that, if the developer of an app has built in a feature that asks the user for location, he must have a good reason for doing so – location-based apps like Foursquare, mapping apps or local media properties sites/apps that are providing geo-aware weather, sports scores or movie times are good examples of apps that can ask for location and with which users tend to share location.
Yet, some exchanges and networks claim 70, 80 or even 100 percent lat/long impressions. How can this be? One more time – 5 to 10% of all mobile impressions have lat/long but a given exchange may claim 80% lat/long? The answer is some people are making up lat/longs. This is the dirty little secret of location-based advertising. About 12 months ago, some publishers figured out that location was the one attribute that really moved the needle in the exchanges and the “inferred” (read “made up”) lat/long was born. If buyers are paying a premium for lat/long location data and it’s the only thing that can give a publisher’s mobile impressions a price boost, manipulators will find a way. And they have. In 12 months’ time, the number of lat/long impressions in exchanges has grown from the industry average 5-10% to the current 70-80%. Even though, generally speaking, lower quality, rather than premium impressions, end up in exchanges, so one would expect that premium lat/long impressions would be used by publishers and their location specialist partners and fewer (not more) lat/long impressions would be making it into the exchanges.
The methods being used to generate “inferred” lat/longs fall into two primary categories (although there are more): “centroids” or “randomized” lat/ longs. Centroids are lat/long coordinates that are generated by software programs that automatically pick the center of a geographic region as a substitute for either no location data – many corrupt lat/longs are dead center in the middle of the country – or for lesser-quality location data – they are in the middle of a state, DMA, city or zip. Randomized lat/longs are generated by software programs that randomly choose lat/longs within a region. This is a really big problem for the evolution of the highly granular location targeting that has so much potential and has marketers so excited. If a marketer is targeting a particular store location, mall or office building, a particular neighborhood or city block or an audience segment that is based on inferences drawn from the context of a user’s location, then bogus location coordinates are almost certainly targeting the wrong place and the wrong people and driving down performance. Garbage in, garbage out.
So what to do about it? At Verve, our roots are in building mobile tools for publishers and we work with many of the largest premium content publishers in the US providing them with a mobile publishing, web and app platform that powers their mobile content distribution. So, for about a third of the publishers we work with, we are the first party technology platform – we know with 100% certainty what location the GPS chip in the device is generating. Also, most of these properties are location-aware, giving users good reason to opt in to sharing location. Further, for the publishers we work with who aren’t on our publishing platform, we work with most of them directly, through first party business development relationships. If we see data that our systems detect as suspect we call them up and rectify the situation — either they send us good data or we don’t work with them. When we do venture into the mobile exchanges we leverage our unique proprietary technology that allows us to recognize and parse ad impressions with genuine location data from those with false location data and cherry pick verified impressions. We work with publishers with over 108M US unique users and see almost 10 billion impressions monthly and we have more verified lat/long impressions than anyone in mobile. And when we use DMA, city, zip, user supplied or IP targeted location data – which can be valuable under the right circumstances and can help advertisers achieve scale in this emerging technology arena — we don’t hide the ball and call it something else. All of these things – cultivating inventory partners with higher than average true lat/long data, working directly with partner publishers so we can have confidence in the data being received and developing proprietary technology and models that allow us to identify and reject “inferred” lat/longs – are important in making sure we can operate our business on legitimate location data and advance location targeting in mobile. But we are also working on a broader solution that could become an industry model – a system that scores every impression for the likely accuracy of its location data and estimates the likely original source (e.g. ‘this centroid is likely derived from this zip code’) so even if the problem of “inferred” lat/longs continues we can provide a measure of transparency.
Location targeting in mobile has enormous promise. With the increasing penetration of smartphones, increasing user migration from online to mobile consumption, the increasing development of location aware apps and the increasing comfort by consumers with sharing location, we will see steady growth in high-quality, location-aware ad impressions sufficient to make this promise a reality. The lack of transparency in location data only sets us back.
(In online exchanges cookies provide the data points that act as the value drivers for impressions (a user who has visited a car site is an ‘auto intender’ and that information is stored in a cookie and that impression commands a high price for an auto marketer), but in the mobile world where we have a fragmented technology landscape with multiple devices, operating systems etc, the third party cookie state is very limited. So in mobile exchanges mobile marketers typically bid on IAB content category (I want someone reading business or mom content), or device type (I want people on high end phones) or location (I want people in lower Manhattan).