Within six months of being on Google Cloud, Motorway entered the Google for Startups Cloud Program. This unlocked access to training, consultancies, and support in the development of its pricing tool, named Real-Time Price Machine (RPM). 

Now the company develops its own ML models on Vertex AI to fine-tune pricing estimations that take into account all the nuances that might affect a vehicle’s valuation. This includes analyzing information provided by the car seller—such as the vehicle’s make, mileage, service history, and defects—coupled with market research on the supply and demand for each vehicle at the moment of valuation. Additionally, Motorway uses Vertex AI endpoints to incorporate real-time data from the previous day’s auction to understand the current interest of the car dealers on its platform and how much that vehicle is likely to achieve in auction.

“Each used car has a unique story and circumstances, making it traditionally difficult to estimate their value with accuracy. But for dealers on our platform, who are working to make a margin on each sale, estimating the price by even just 1% too high or too low can have a huge impact. To tackle this, we use advanced machine learning techniques that consider the broader market, individual vehicle details, and our own marketplace dynamics, allowing us to provide accurate valuations and ensuring the best possible outcome for both our sellers and buyers,” explains Ben Jones, Head of AI & ML at Motorway.

As an AI/ML expert, Jones notes that pushing ML models to production is traditionally challenging. In fact, it’s common for companies to spend multiple months working on something that often struggles to see the light of day. “It’s commonly known in the industry that typically fewer than 20% of machine learning models make it to production,” he says. However, that’s not the case at Motorway. Since establishing its current technology and processes, more than 80% of Motorway’s ML projects make it to production, which means Jones’ team can launch new customer-facing features faster, and keep current ones up to date and fully optimized. 

And speaking of new customer features, Motorway is currently developing the next iteration of its Intelligent Vehicle Discovery tools for car dealers using its platform. This solves a challenge that’s particular to Motorway, compared to other online marketplaces: “Unlike ecommerce websites that list recurring products, our auction platform has entirely new stock every day. Since we never see the same vehicle twice, it’s difficult to use traditional recommendation techniques to learn what is useful and relevant to a user based on their interactions with vehicles listed on our page,” Jones explains. 

To solve this, his team is developing a bespoke personalisation engine to help car dealers find vehicles that match their interests by using criteria such as mileage, condition, age, and price, as well as make-model, on the Motorway platform. Additionally, more recommendation features are on the horizon for Motorway users, such as the ability to browse through similar, related vehicles to the ones they’re interested in, across similar makes or manufacturers.

To provide this experience for its platform users, Motorway is exploring Cloud Functions and Batch for training Motorway’s ML models, looking across millions of user interactions to identify which users are likely to click on which vehicles. The team runs the results with Cloud Scheduler and uses Pub/Sub to identify when these matches are complete.

Artifact Registry and BigQuery store all that data. Meanwhile, another vector-specific database provisioned by Google Cloud stores vehicle vectors used to identify which vehicles are similar to one another according to relevant criteria for car dealers. “With generative AI, we augment those vectors using embeddings from LLMs to really capture dealers’ interests and how they’re interacting with different kinds of vehicles. Finally, we then use Gemini to quantify the performance of our model for each of our dealers at scale,” Jones explains.

With well-established internal MLOps capabilities, or as Jones puts it, “the ability to do things better and faster,” Motorway is looking forward to launching even more exciting technologies, transforming the way we all buy and sell used cars.