Modern technology allows marketers to know a lot about consumers, right down to their precise location. But how accurate is most of the technology? GroupM’s Timothy Whitfield puts the technology to the test.
Over the past few months I’ve been approached by many adtech vendors who are selling Hyperlocal Geo Targeting for mobile inventory. I’ve been suspicious of the sales people when they say things like: “We can target users with an ad for McDonald’s whilst they are standing outside of KFC.”
This all sounded a bit too good to be true so I decided to look ‘under the hood’ and put them to the test.
The purpose of the test was to check the accuracy of the location-based services that each of these adtech vendors can offer. Most of these companies claim that they can determine the location of an individual user to a resolution of circa five meters from their actual location.
It was important to me that the test be fair so I asked each adtech vendor what information they needed from me. Some companies required my IP address, some wanted me to download specific apps and others wanted me to activate Location Services and then my IDFA (Identifier for Advertisers) and even my UDID (Unique Device ID).
I created a ‘digital information pack’ about myself. I collected my UDID, IDFA, IP Address whilst at home and work and whilst on 4G versus wi-fi. I downloaded a number of apps like TuneIn Radio, Relax Melodies, Words with Friends etc… I ensured that Location Services was enabled on my phone and that I was sharing my location with each of these apps. I used these apps at regular times during the day as well as keeping the apps open in the background.
I sent all this information equally to each adtech vendor.
After one week of regular app usage I asked each of the adtech vendors what information they had about me. I don’t want to name the vendors by name but I can happily share the results:
- Vendor 1 – Couldn’t find me
- Vendor 2 – Nothing
- Vendor 3 – I downloaded five mobile apps that this vendor had interaction via Software Development Kit (SDK). I used the apps multiple times perday for almost a week. They were able to find my home address whilst my phone was turned on and I was asleep but there were not able to find mewhilst I was at work or “out and about”.
- Vendor 4 – Nada
- Vendor 5 – Tracked my geo-location accurately to where I was standing (circa five meters). Also identified my home IP (but was 10 kilometres off from where I really was) but could track my online habits and that of other users in my household.
This was certainly a ‘sub-optimal’ situation. These adtech vendors pride themselves on accurate location targeting and only one of the five could find me.
The other four had no idea where I was even after I had shared with them my ‘digital information pack’. This felt like I was standing on sidewalk waiting for a friend to meet me with a huge flashing neon sign above my head saying ‘here I am’ and my friend walking right by me without seeing me.
As a media brand you need to understand more about how mobile location based targeting works. So let’s have a quick overview of the methodologies used in hyperlocal geo targeting.
Concept 1 – Location Based Services
This is where the Mobile Phone receives a ‘ping’ from multiple satellites that are orbiting the earth. Each satellite sends out a very basic signal at regular intervals. Specifically, each satellite broadcasts a continuous signal containing the time-stamp and unique satellite ID.
The exact position of these satellites are known as they are in a geostationary orbit. Remember from high school maths that speed equals distance over time.
The speed is constant at the speed of light. The time is calculated by the difference in microseconds between when the ‘ping’ is sent from the satellite and when it’s received by the phone. The only variable to solve for is the distance.
If you can get a ‘ping’ from three (or more) satellites then you can use trilateration (not triangulation, boy it bugs me when sales people get that wrong) to calculate the location of the phone.
All these numbers are very accurate. The speed of light is known to 10 significant figures. The location of the satellites to the nearest meter. The time-stamp in the satellite ‘ping’ is to the nearest microsecond.
Generally speaking the current resolution is about five meters. It’s technically possible to be more accurate but I’ve heard that most governments feel that it’s a security risk to give any more targeting granularity to civilians. However, I’ve heard that the US military uses its own satellite system which is accurate to about 20cm.
Media brands should be thinking about how this data is used by adtech vendors for mobile campaigns. When the user activates Location Services the operating system calculates the location and stores it in memory as a Latitude and Longitude. If the user shares Location Services with a specific app then the developers of that app can pick up the Lat/Long and pass it to the ad call.
However, the ad-call needs to be designed to accept Lat/Long information. The most common standard to accept mobile Lat/Long information is called OpenRTB. To pass the data to OpenRTB the web developer needs to call a supply side platform (SSP) via a software development kit (SDK).
Phew! That was a long description but I hope you see how difficult it is to get valid location based information. It requires the following…
Mobile App Inventory – It doesn’t work for Mobile Web Inventory! Mobile Web inventory doesn’t communicate with the Operating System of the phone.
- Location Based Services On – It only works if the user has enabled location based services. This is ON by default but many people often turn this off.
- Location Based Services Shared – It only works if the Location information is shared with the App. This is OFF by default and the user would need to opt-in to turn it on
- Open RTB – The inventory would need to have valid Lat / Long information passed into the bid request.
I’m seeing only about 25% of the total number of ad-requests coming into the network having the combination above. That means that 75% of the mobile impressions you are seeing either DON’T have factual data or they have used some algorithm to extrapolate and guess what the location may be.
Concept 2 – Wi-Fi Mapping
The second concept to get your head around is the wi-fi mapping. When you sit down at a coffee shop you may notice that they have a free wi-fi service.
Whilst you order your morning coffee you may decide to switch from 4G to wi-fi to get free internet access. When you switch this will create two records in the wi-fi mapping tables at Google & Apple.
The first record is based on your satellite-based address as per the method above. The second record is your wi-fi address which you switch to only moments after arriving. The phone manufacturers can de-duplicate these two data points together and accurately determine the geo location of that Coffee Shop’s free wi-fi router. They only need a few data points in order to accurately identify the Lat/Long of that wi-fi router.
When new users log-on to that same wi-fi router then the Operating System can automatically determine your Lat/Long without needing to use the satellite data.
This method is just as accurate as Concept 1 – Location Based Services, however, it requires the same services to be enabled on your phone in order for it to be used.
Concept 3 – IP Address
The only time you will ever see someone talking about the IP address from the cell tower and telling you that it’s a valid signal for marketing is in a sales deck.
From an engineering point of view the IP address from cell towers is not a valid signal. IP addresses are centrally governed by organisations like APNIC. These companies give out groups of IP addresses into classes.
A typical range for an organisation to buy may be a class C range, which will have 254 usable addresses.
Have you ever been at a concert or an event and struggled to make a call or get online? That’s because each cell tower can only support a certain number of connected devices and a maximum bandwidth.
Testing hyper local mapping.
This relates, in part, to the number active IP addresses assigned to any tower.
When you drive to work on the 3G/4G network you will no doubt drive/travel past many different cell towers. Each time you do then you will be given a different IP address. If the cell towers are around 10km apart in a city then the minimum resolution that IP Address data could determine would be 10km.
This map shows where I really am versus where Mobile Location would put me based on IP Address. It thinks that I’m about 2.5km away from where I really am.
In summary: As a client, if you are thinking of using hyperlocal geo make sure that you understand the accuracy of the data.
Here’s my top three must-dos:
- Make sure that you’re selecting a vendor that can get you access to accurate satellite or wi-fi location-based service data;
- Check where the ads are being served (in-app vs. mobile web), as in-app supports passing a unique device ID for re-targeting, but mobile web doesn’t.
- Ask if the targeting that you are receiving is deterministic (25% of the data) or just probabilistic (75% of the data).
Hyperlocal Geo Targeting is a relatively new and exciting addition to a marketer’s toolkit, and can produce fantastic campaign results.
Make sure you’re getting the best product in market and you’ll make the most of your mobile advertising budget!
Timothy Whitfield is the director of technical operations at GroupM.