Starbucks is on the corner of every major street in every major city, it seems. According to data aggregated by location analysis startup Placed, it is the third most popular company in the country. However, if you place a Dunkin Donuts in the same market as a Starbucks, the New England doughnut shop attracts three times as many customers.
Last week, we caught up with David Shim, CEO of Placed, the aforementioned company responsible for the intriguing Dunkin Donuts vs. Starbucks insight. While location analytics could prove helpful to marketers looking to continually increase the personalization of their marketing campaigns, it also harbors some hidden challenges, not the least of which is the unreliability of GPS technology in finding a person’s precise location.
According to Shim and Placed Head Engineer Nick Gerner, GPS systems and Wi-Fi triangulation are surprisingly imprecise. “We’ve been hearing about how GPS has five-meter resolution or whatever for the last 20 years,” Gerner said. “When GPS first came out, it was going to revolutionize everything because you know exactly where you are and that’s not true.”
For someone using something like Google Maps while driving down a highway or along the outskirts of a medium-sized city like Raleigh, a phone’s GPS does a good enough job.
However, some of the more interesting insight for marketers is found in large cities in areas where there is a large concentration of retailers. It is in those areas where pinpointing one’s exact location is markedly difficult. “In a place like an urban jungle,” said Shim, “GPS can’t get a fix because you’ve got tall buildings all around you. Or if you rely on Wi-Fi triangulation, the accuracy of Wi-Fi is less than GPS. Wi-Fi gets you within 20-30 meters in an ideal situation, and 20-30 meters in downtown New York or San Francisco or even Seattle, that could be thousands of businesses. You’re talking about four or five blocks in the city.”
According to Chim, this lack of accuracy can be typified by location check-in apps such as Foursquare, where the restaurant or store where someone is standing shows up in the top ten but often as the sixth or seventh choice instead of the top choice. “In our tests that we run,” Chim said, “we found that 90percent of the time, if you take a latitude and longitude collected from a smartphone device to find the closest location, you’re going to get a wrong answer.”
GPS and Wi-fi data are therefore insufficient. Placed relies on a machine-learning model that incorporates customer feedback data from about 30,000 volunteers. These volunteers are sometimes given incentives (Chim mentioned a Walmart program which gives away $5 gift cards in exchange for the ability to track and collect an individual’s location data) but for the most part are interested in being fed marketing information more relevant to them.
Placed takes this data, which consists of location information, application use data, and even seemingly minor information like battery life, to form models that correlate those variables and attempt to predict consumer behavior. They are currently in the process of building up their databases and models while finding a little insight along the way. “We actually have a training set of data that we built models against,” Chim said. “To give you an idea, we’re at 2 million plus validation points… We actually have a feedback loop from consumers that says, ‘Hey, this is when I went to Walmart and I was here during this time.”
The model takes contextual information into consideration, with Chim noting an example where an individual is wandering around downtown San Francisco at one in the morning. The location data may show that he is near Banana Republic. However, unless he is a criminal or a worker who left their watch in the store, he is unlikely to actually be inside the store. Instead, he is significantly more likely to interact with the adjacent bar that is open until 2am, Chim noted.
From a market perspective, location data and analysis can prove quite valuable. For instance, someone who remembers to fill their prescription may just remember randomly or they may have been triggered by a drugstore they were near. A drugstore app could remember this information and remind the user when he is near that location again.
To cite another example, an Amazon barcode scanner app could aggregate information on how often, and more importantly where people are searching for an electronics system like a PS3. “What becomes valuable is,” Chim said “if you’re someone like Amazon, you can go in and say, ‘look, 25 percent of scans using the Amazon scanning app occurred at Best Buy. Fifteen percent occurred at Walmart, six percent at Target.’ So it gives a sense of why someone would open up the app.”
Chim even appealed to the journalist he was talking to, saying that publications might be well-served in tracking the location data of the people reading their articles, with potentially a high volume of page views occurring in coffee shops.
Speaking of coffee shops, few examples better represent the surprises that can come out of this type of analytics than Dunkin Donuts vs. Starbucks. As Placed continues to build its database and machine-learning model, they hope to find many more interesting and useful tidbits of enterprise information.