Client introduction
AAA AUTO is one of the biggest used car dealerships in Central Europe, with more than 30 years of experience with used cars, operations in 5 countries and over 2000 professionally trained employees. They sell over 70,000 cars each year, and a substantial part of these trades originates on their website.
“For a long time, we have been concerned about the sales conversion of users who visit our online vehicle sales portal. At the same time, we needed to reduce the time to sell less sought-after vehicles. An innovative solution that, based on the knowledge of the potential customer's behavior, offers a tailored price to each of them, significantly increasing the likelihood of genuine interest in other car from our available stock.“
Business Case
The web portal for the presentation of the offered vehicles was created primarily to generate new leads for individual branches in the region, to create interest to physically visit a particular car and decide to purchase a selected or similar vehicle within the search criteria. Despite high traffic, optimized pages and detailed analysis of individual user behavior, the portal did not generate the expected number of new leads for sales. The second challenge was to increase the likelihood of sales of stocked vehicles, which had long been of lower interest.
Solution
To enable this whole project, we start off with a machine learning model focusing on web visitor‘s propensity to lead. That is the key KPI we aim to increase.
The model predicts the probability to lead of each visitor within a certain time frame helping us focus on the visitors that otherwise would not send a lead and later would leave the website. That was our target group.
To visitors in this group is offered an instant unique proposition, a limited-time discount voucher on certain cars, requiring no further haggling with the on-site salesforce. This discount is guaranteed if used by the same person for the same designated car.
Cars that are eligible for this type of discount are carefully selected by a model. We aim is to promote mainly ones that receive no or very low traffic or ones that gather minimum leads in a certain timeframe.
Technical implementation
We used Python for analytics, data wrangling, job scheduling, training and serving the ML/AI model; then Redis for lightning-fast in-memory key-value storage; Exponea/Matomo + CosmosDB + Azure Storage Queue, Synapse was used for website traffic tracking and customer statistics aggregation; PowerBI for reporting and results tracking.
Captured benefits
There are multiple benefits captured implementing the personalized discount solution.
As a direct result of offered vouchers, more people tend to send the reservation forms. Currently 50% more leads each month then the client had before.
Because the customers with discount vouchers are determined to use it, their conversion to deal is significantly higher. This solution help to increase conversion rate by 38% from lead to deal, saving significant time of call center agents.
