Going to Beta Testing
To be certain, nothing that I had crafted could be considered the most elegant or safest solution, but it did complete the prototype setup, enough that we could install the sensors in the store. You can see it deployed in Figure 9-4.
Figure 9-4. Testing the sensor network in the studio
The initial beta installation went smoothly; we had the sensors up and running and they began to immediately collect data. Over the first week, we made no effort to visualize the data, but I recieved a call from the store owners saying that just looking at the raw totals they had already identified a bad flaw in the layout of the store. When the store became slightly busy, there was a bottleneck that formed in the middle of the floor that then caused one-third of the store to be essentially cut off. The owners are there six days a week, and the store had been open for years, yet they had never noticed this behavior until they actually saw the data. They immediately set about rearranging the store layout to correct it. Something else happened the first week: people in the store began to notice the sensors. At that stage in the project, the sensors were bare PCB boards with wires and a battery sticking out of them with a small red LED power indicator light. This was right after the horrific bombing of the Boston Marathon, and people were genuinely alarmed by these crude-looking little devices they were discovering all about the store. I received a call one weekend from the store owners asking me to do something after three people approached them to let them know that they found a bomb under one of their display tables (see Figure 9-5) and were getting as far away from the store as possible. We learned the first big lesson from beta testing: the sensors must look like they belong in the space. We immediately obtained a Makerbot and very quickly designed and printed cases for the sensors.
Figure 9-5. A bare sensor taped under a display table in Tigertree
We had the first beta round up for a couple weeks. We gathered a good amount of data and began working on creating meaningful tools out of this new information source. We had originally called the sensors “people counters” because, well... that’s exactly what they do; they count people as they walk by, but when you looked at the raw data and saw the clumps of activity, something didn’t seem to line up. We ended up sitting in the store for long periods of time and observing what was happening and comparing it to our data. People didn’t just linearly walk around the store; they would go to one part of it and meander back and forth multiple times or stay in front of one display for a while and move back and forth, each time setting off the sensors. Our sensors weren’t people sensors, they were activity sensors. Fifty different people didn’t walk by one sensor in five minutes; one person was concentrated in that single area for those five minutes and triggered the same sensor fifty times. We realized that we needed to recontextualize the sensor data, and to more accurately convey what was actually happening in the store, we needed to know exactly how many people were in the store at the time of the activity.