They're not really counting cars, they do "parking lot fill rate analysis". Much easier to do - you just subtract the percentage of the parking lot that has the same color as the ground.
I see two main issues with this analysis. First, the amount of profit per customer visit is not necessarily constant, and can depend on things such as gas prices (e.g., cheap gas encourages two shopping trips per week, even though the total amount spent on groceries remains the same; expensive gas makes some people take the bus).
Second, is even the amount of visits well estimated? The article mentions things such as 0.7% accuracy in profit estimation. Random fluctuations in the number of car passengers, time of visit etc. (how often are these images taken anyways: once a month, or ten times an hour? the article doesn't mention) can lead to much greater variations than that.
> First, the amount of profit per customer visit is not necessarily constant, and can depend on things such as gas prices
Given the scope of the company doing this kind of analysis, I don’t think variance is going to be significant. Given a large enough sample or pool of data, some kind of “regression to the mean” effect is going to kick in. Remember it is all “estimation” in the end anyways, and I can see the value in having some kind of observable data as opposed to data based on “honest” reporting by those you are trying to study.
I see two main issues with this analysis. First, the amount of profit per customer visit is not necessarily constant, and can depend on things such as gas prices (e.g., cheap gas encourages two shopping trips per week, even though the total amount spent on groceries remains the same; expensive gas makes some people take the bus).
Second, is even the amount of visits well estimated? The article mentions things such as 0.7% accuracy in profit estimation. Random fluctuations in the number of car passengers, time of visit etc. (how often are these images taken anyways: once a month, or ten times an hour? the article doesn't mention) can lead to much greater variations than that.