How to overengineer a car purchase

Prices are one dimensional, but consumer preferences are heterogenous and multi-dimensional. It follows then that wherever there’s uniform pricing, individual preferences are being smoothed out. This in turn creates pricing imbalances and opportunities for consumer surplus. In 2024 I had a preference to buy a Tesla. At the time, their used car marketplace had reasonably competitive pricing and more transparency than other sellers (e.g. on what features were included with the car).

After watching the market for a couple days I noticed a few things:

1. The price of each car went every day or so until there was a sale (also known as a Dutch auction)
2. The value that Tesla and/or the market put on features diverged with mine. For example, FSD might increase the cost by $3000 but I consider it to be worth $4000. Conversely, it may value the performance version of a car as $4000 while to me it adds nothing.
3. They have a publicly accessible API

This meant that I could:

1. Price out how I valued all the features
2. Write a script to score different cars based on my value set
3. Run it periodically and watch for good deals

The script that I wrote¹ sends a text whenever there’s a new car at the top of the leaderboard or when the top car drops in price. In writing it I also learned that there were important features exposed (or inferable) through the API but not surfaced on Tesla’s website, e.g. the hardware version of the car.

After running it for a couple weeks I noticed that the good deals were getting fewer and farther between, so I waited for the next good one and (after waiting one more day) got the car! The final day’s price drop of $300 was enough on its own to make the effort worthwhile, but the savings on the car using my personal heuristics was around $2000 vs the average.

The lesson here applies more broadly, especially to large purchases in big markets like housing. Unfortunately, it produces an incentive for some markets to lock down programmatic access to their data. The cat and mouse game is likely to escalate as AI agents start doing more shopping and sellers lean more into dynamic or targeted pricing.