The sales prediction system based on Machine Learning and AI is born from the need to analyze sales performance and use the most cutting-edge technology to know in advance whether a promotional campaign or a particular product will have good sales performance.
At BertIA, we have defined the following objectives to create a good product:
- Ability to predict with acceptable accuracy for a prediction model the performance of a product and/or promotion at a given time.
- Provide the client with relevant information about the model’s construction and the importance of each parameter (EDA) in order to understand the result of the predictions without having knowledge of Machine Learning.
- Create a powerful prediction tool that is also intuitive and easy to use, so that anyone in the company can easily use it.
The application created by BertIA uses the most advanced technology in terms of Machine Learning to predict the performance of marketing campaigns for the client. This application uses all the resources provided by Microsoft, thanks in part to the Azure Machine Learning Studio resource.
The process of using the application is very simple: the user enters the product data to be predicted along with other relevant data. The program then returns information about the expected performance of the promotion. This prediction is supported by historical data from the client company with more than 5 million transactions, making the predictions completely customized for this business.
By having the application hosted in the Azure cloud and using the Machine Learning Studio resource, it facilitates monitoring the performance of the model. In addition, it offers the option to retrain the model whenever desired or simply schedule an automatic update. This helps the client not notice any drop in the model’s performance or updates, as the same application will continue to be used without making changes to the architecture; everything will be managed from the Machine Learning Studio.
This ML application created by BertIA is a SaaS solution that allows for both an initial payment business model for the project plus a subsequent period of maintenance and supervised retraining, as well as a flat-rate modality where the client pays a subscription as they use the service. In this case, the first option has been chosen, but since the application is so customizable to any type of client and business, it could be considered changing the business model in the future, depending on the client’s needs.
The project is still in the testing phase, but the initial results offered by the application are very promising. In an early stage of development, the predictions have a +65% accuracy in the testing phase. Considering that the model is still being refined and fed with new and better-quality data, the final result will have very good performance.
Currently, the service is implemented in this client in a testing and trial phase while the model is being improved. In parallel, all alternatives are being studied to offer this service to any company in the commercial sector, without losing customization according to the needs and characteristics of the company using this service.