Nowadays, because used car buyers prefer to play safe, vehicle history check services are extremely popular. Retrieving car history data by yourself is time-consuming and, in most cases, it won’t even be possible. That’s where VIN decoders kick in. However, due to mismatches and unstable databases, many providers of VIN decoding offer sketchy and unstable history reports. That’s why the team of experts at carVertical set a goal to kickstart the world’s vehicle data economy from the ground up by building an innovative VIN decoder.
Good old VIN decoder – why do we want to change it?
To understand how traditional VIN decoders work, recall a phone book from the olden days. If anyone knew your full name, it could be tied to your phone number and home address via a phone book. However, there were no entries without names. It was necessary to have a name to get access to other data about you. The same goes for some VIN decoders: when the VIN is absent from the provider’s database, data tied to it is also absent.
This leads to a common problem with history reports – unstable data. With unstable data, some information is often missing or even incorrect: for example, the system might hold the wrong color or false engine specs, etc. Consequently, clients often buy car history reports from multiple providers to get as much available information as possible. Imagine how expensive and time-consuming this is.
Instead, our team of professionals at carVertical constantly gathers data from multiple sources and vendors. Due to this high-maintenance solution, carVertical car history reports always offer accurate and up-to-date information.
AI-based VIN decoder
The amount of data available differs by provider. Additionally, VINs have multiple standards, which vary based on the period, region and manufacturer, etc. Instead of putting enormous effort into decoding all possible variations of the VIN, we take an alternative approach – our innovative VIN decoding algorithm determines all possible patterns, even patterns that we might not even know exist. Here’s how it happened.
We made a hypothesis that VINs have a logical sequence of characters, which could provide meaningful information about a vehicle when used in the correct combination. By ‘information’, we’re talking about model name, fuel type, and body type, etc. We then used TensorFlow to build our neural-network-based algorithm for the automotive industry.
By testing multiple model architectures, we managed to reach significant accuracy with our algorithm, which sparked our passion and led to an AI-based in-house VIN decoder solution. We built an entire framework dedicated to preparing the data and training multiple models. The framework has a feedback loop with a specific job – to pick the best performing models. Thus, we ended up with an AI solution that is capable of continuous learning and self-adjusting to the benefit of those requesting car history reports.
The dataset of hundreds of thousands of entries, combined with a deep learning model, helps us make accurate predictions about millions of vehicles. Changing VIN patterns are quickly picked up by the algorithm, the number of extracted vehicle features increases and we are not limited to external VIN decoders. This was a true game-changer.
We are working towards a transparent used car industry
If we are going to make the used car industry a secure and trustworthy place, we can’t stop here. The carVertical data science team is constantly polishing and refining our VIN decoder, implementing new features, and solving emerging issues quickly.
The behaviors and preferences of used car buyers, sellers, and data providers also play an important role. The more we all take the quality and legitimacy of vehicle history data seriously, the sooner problems from outdated VIN decoders will become a bitter past.