More on Starbucks, Microsoft and Geocoding
I got an early start today; after all it was going to be good, old fashioned field work. I decided that I needed to take a look at the locations of the Starbucks shops I mentioned in my last blog (above). I wanted to find out how well the map matched reality.
I prepped the “Geotagger One” and tuned up the equipment for the mission (see in the attached photo) at our Secure Sensing Facility in Laguna Hills, CA).
Pretty impressive, huh? Ok, so I don’t really have the camera arrays and sensor equipped vehicles, or even the technology of a Microsoft, Google, Navteq or TeleAtlas. At least I had the wherewithal to get out there and see if their technology had been put to good use. I had a simple digital camera, an old TomTom, a pencil and a pad of paper. Oh, I had the map from Starbucks, a couple from Google and print Thomas Guide, just in case. Unfortunately, I did better at location the Starbucks shops than Microsoft/Starbucks or Google. I suppose everyone will want to know my proprietary techniques, so read on.
I prepared a route that would take me by eight of the locations on the map. The locations were selected by my based only on the rationale that they were along a reasonable path and that eight seemed like it would be enough to give me an indicator of the veracity of the location data.
I managed to find all of the Starbucks, but most of them were not located where they were shown on the map. I pulled up and parked the GeoTagger One in front of each Starbucks shop, checked the address (all were correct) and took a GPS fix no more than 10 feet from the front of the store.
Since pictures are so useful, I prepared a set of maps summarizing the results for each of the eight locations (one map for each location). The squatty dark green markers on the maps are where Starbucks and Microsoft think the stores are located. The red pins show where the shops are actually located. The green pins show the cross streets that were provided in the address information that accompanies the maps on the Starbuck website.
By the way, if the distance errors were small, I estimated them from the map. If they were large, the error distance is the distance between the locations when driving (calculated by running a route on Google).
What might we conclude from this effort? At first blush it appears that Starbucks knows the addresses of its stores and the cross streets nearest to the stores. Unfortunately, it appears that Microsoft has some significant weaknesses when it comes to geocoding. About two years ago, I was told that they were converting to a rooftop geocoding source, but that appears untrue, or at least they have not reached parts of Orange County, California yet. Worse yet, it is clear that Msoft did not use the cross streets in their geocoding/mapping process.
It appears that cross streets are more accurate than whatever sources or techniques Microsoft is using and, on the average, equal to the tools and techniques employed by Google, although this may be the influence of one significant outlier. Of course, my method is extremely inefficient. Hmmm…maybe I will write about how to solve that in the future!
Google did quite well, but there were several troubling aspects to Google’s performance. In several cases, there were two close together symbols on the online map representing potential locations of Starbucks. One symbol was the familiar Google Plop, while the other was a red dot that indicated a location within the same category. When I clicked on these dots to interrogate it, I found it to be the same location as the Plop. And then the plop move to the dot and mated during my next pan of the image. Not sure what to make of that, but it seems that the data somehow merged on the fly. Interesting.
Of course, Google is now showing unverified locations on their maps. Isn’t that cute? What a bad idea!
Take a look at this image, made by pasting together two version of the same area. As you know, you cannot have two information windows open at the same time. So, I opened them sequentially, captured them and merged them together. The unverified dot indicates a malformed location that is actually several miles south and shown in its exact location in the bottom window.
Well, Google, you won the sweepstakes, but were disqualified on a blunder. I thought Google did a lot of data mining and data harmonizing. In the location where the Google error was greatest, the accompanying photo of the Starbucks location was a field of grass and brush.
Well, that’s enough for today. I need to finish up a report, so I can cover the taxes that I will owe tomorrow. One important lesson from today’s blog – don’t use Local Search to find the location of a tax accountant. You might be looking for the storefront long after the post office is closed.
I will delve just a little deeper into geocoding errors/content errors/map errors and how they impact Local Search over the next few blogs. These positioning problems are not going away if we continue to use the same old methods. We need to think about something new.