Astara is a vehicle import company. In Colombia, it exclusively represents Dodge, Fiat, Hyundai, Jeep, Opel, Peugeot, RAM and Volvo.
Astara had the need to have a forecast of the number of vehicle registrations in each of its markets in order to accurately determine the number of vehicles for import.
They made the prediction of 2 of their markets in tools such as excel and based on the knowledge of the automotive industry, however the COVID-19 pandemic created greater uncertainty and a decrease in the assertiveness of the forecasts, they wanted to include machine learning in their solution and that it take into account the impact of the year 2020.
Their priority was to improve the accuracy of existing forecasts and have new scenarios for the rest of their markets, segments and brands, a total of 7 scenarios or use cases within Amazon Forecast, additionally they needed support in the analysis and creation of a new one. variable that would allow evaluating the impact of the COVID-19 pandemic (year 2020) on vehicle registration records, we discussed with the client the benefits of AWS, the ease of integration, the low cost and the benefits of having a scalable architecture , benefiting from economies of scale and interaction between AWS services.
Lambda functions were created in Python language to upload Excel files and SQL Server tables to AWS. For the data transformation and creation of datasets for Amazon Forecast input, a lambda function was created that stores the datasets in an S3 bucket. Jobs and Workflows were created in Glue to automate the implementation of each of the steps within Amazon Forecast to achieve the prediction of vehicle registration in Colombia in the coming months.
Within the Astara server, the results visualization board was made in PowerBI, taking the resulting tables in Amazon Athena as a source.
Reduce the analysis time of the behavior of markets and macroeconomic variables, from 2 weeks to approximately 12 hours.
Have the registration forecast for 5 types of markets, segment groups and brands.
Go from predicting 6 months to more than 50 months.
Automate the creation of data sets and monthly forecast, and visualization of results on the PowerBI dashboard.
Improve the forecast between 5 and 10% in the precision of the results returned.