posted: July 15, 2023
tl;dr: A case study of how software automates aspects of retail automotive industry jobs...
Prior to my recent vehicular transaction, the last time I sold a car was in 2018 when I sold Carmax an Acura MDX with 150,000 miles on it for more than I was expecting to get. Carmax’s process in 2018 took about an hour, most of which I spent waiting while Carmax technicians inspected the vehicle. Presumably they were going over a checklist to see what features it had and didn’t have, to see what worked and didn’t work, and to estimate the amount of cleaning and repair work that might be needed to get it into a sellable state. Since I keep my cars in better condition than the average owner, I believed this process benefited me, and contributed to the higher-than-expected offer they made me.
Fast forward to 2023, when I wanted to unload an Acura RDX so that I could buy a 2023 Jeep Wrangler Willys 4xe 4-door. Living in the Valley of the Sun, I am well aware of Carvana, an online car retailer which is headquartered in Tempe, not too far from the office where I used to work. I knew they had an all-online process for selling them a used car. So I went to their website and filled out a form that asked a few questions about the vehicle, the most important of which appeared to be the VIN, which they obviously used to retrieve any public records about the history of the vehicle. Perhaps they also did some checking on me, to see if I had a history of having committed fraud. I quickly received their offer via email, and all I needed to do to receive my check was to accept it on the website and make an appointment for a Carvana tow truck to take the vehicle away.
I was not impressed with Carvana’s process. They obviously had not inspected my vehicle. They didn’t know that I had just spent over $1,000 to put new tires on it. They didn’t know that there were no major dings or scratches on it, although they also didn’t realize that the lens to the backup camera had cracked, which distorted the image. Clearly they were just relying upon algorithms that used the information I had provided to attempt to determine the market price, and were ignoring individual variability.
So I made an appointment at Carmax, figuring they could easily beat the Carvana offer. I thought Carvana had lowballed me by not paying any attention to the actual condition of the vehicle, other than a self-declared “excellent”, “good”, or “poor” condition question. Knowing that the Carmax process took an hour, I brought along a book to entertain myself. When I got there, however, I was first directed to a computer where I filled out an online form that asked the same questions that Carvana’s had. A Carmax representative did take the keys from me, and someone drove the car away from where I had parked it. But within a few minutes the rep presented me with their offer. There hadn’t been enough time to do any sort of detailed inspection. I had read only three pages in my book. I asked the rep what they had done when the vehicle moved, and he said that he drove it a little bit to see the odometer add one mile, so that they knew the odometer wasn’t broken. He had ignored the receipt I left in the passenger seat for the recent tire purchase.
Much to my surprise, the Carmax offer came in below Carvana’s. Clearly the Carmax process had changed from five years ago, and they, like Carvana, were now relying upon algorithms to appraise vehicles. I wondered what had happened to the technicians who used to work at Carmax inspecting vehicles all day. There did seem to be fewer workers in the Carmax building than I remember.
What I witnessed was algorithms taking over the car appraisal process. New companies like Carvana, and older companies like Carmax, have decided that algorithms do a good enough job. It no longer makes economic sense to pay technicians to inspect individual vehicles and adjust some offers up and other offers down. Technicians cost money, and the companies have decided that the return isn’t worth the investment. Better to let algorithms do it: they ignore the actual state of the vehicle, aside from public records such as accident history. As a result, sometimes they overpay for worse-than-average vehicles while other times they underpay for better-than-average vehicles. Unless many of the owners of better-than-average vehicles go elsewhere, it all balances out in the end.
I ended up using the Carvana offer as a negotiating point to up the value of the vehicle when I was negotiating with the Jeep dealer to trade it in, which I ended up doing. The Jeep dealer didn’t do a detailed inspection of the vehicle either; the dealer no doubt also used algorithms. Arizona has a consumer-friendly aspect of the vehicular sales tax: when you trade in a vehicle for a new one, the sales tax is owed not on the entire price of the new vehicle, but on the net price after the value of the trade-in is subtracted. The Jeep dealer gave me a higher value than Carvana’s offer, plus I got the benefit of reducing the sales tax due on the new Jeep.
Software continues its relentless march through the economy, rendering more jobs obsolete. In the past five years it appears to have done so with car appraisers. I wonder to what degree the transition to purely algorithm-based appraisals is also happening in the real estate business?
Related post: The app will take your order now
Related post: The algorithms will determine your price now
Related post: The self-driving car will take you to your destination now