Auto auction app uses machine learning to aid inspections

TradeRev-H, an enhanced version of ADESA's TradeRev app for one-hour online auctions, seeks to tap machine learning to change how vehicle inspections are done.

KAR Auction Services Inc. unveiled the enhancements to TradeRev in March before the NADA Show.

To indicate that the technology is groundbreaking, the "H" is for Grace Hopper, a U.S. Navy rear admiral credited with developing the precursor to the Common Business Oriented Language, or COBOL, computer- programming language.

"For any digital remarketing channel, the visual condition report is key," said Peter Kelly, KAR's chief technology officer. Until now, that often has meant the use of third-party inspectors. TradeRev-H seeks to change that.

It can take video from a vehicle walkaround, automatically create still photos showing parts of the vehicle and upload them in standard condition-report order and format.

In addition, KAR is working to add damage detection. If a scratch or dent is large enough to matter, the video will zoom in for a closer look.

After that, KAR hopes to be able to compare the extent of the damage against its database of damage costs, drawn from its Insurance Auto Auctions salvage-auction business, and factor that into the suggested pricing. Using machine learning, it seeks to predict the exact impact on auction prices of a major dent for a Toyota Camry vs. a Mini Cooper vs. a Mercedes-Benz E class.

"The methodologies don't dictate the use case," Kelly said. KAR's AutoVIN inspectors are crowdsourcing the images to feed TradeRev-H, but the technology can be adopted by other KAR subsidiaries and clients as needed. In the future, he said, KAR might offer differing grades of the app: "a simple inspection a lessee could do, and a slightly more complex inspection that a dealer might do and a third for a certified vehicle inspector — all off the same technology set, using the same capabilities."