Machine Learning: a new forecasting buyer behaviour framework
A new predictive machine learning model
During the last year, our Data Management & Analytical Processing team has worked on machine learning models, based on neural networks capable of delivering highly reliable predictions on a broad range of customers spending behaviors.
In such a scenario as the one we are living today, where
credit and debit card transactions are growing rapidly every day, the ability to analyze and exploit this information can make a different. The analysis of these data can, in fact, provide important insights to companies which are willing to deliver more personalized and efficient services for their customers.
Will the customer make other transactions?
In which market sector?
In which shops?
How much money?
And in which days of the month?
These are the questions that our predictive models can answer, with accuracy up to 99%.
These results were only possible thanks to the latest techniques in machine learning (probabilistic neural networks) altogether with our expertise in financial data processing and a careful study of the problem.
Our models can provide an additional parameter compared to traditional neural networks: a level of the risk – the uncertainty – that the prediction carries with it. This is very useful if we want to provide personalized but also highly reliable services.
More in detail, our framework is capable of:
1. Predicting the probability that a certain customer will make at least a transaction in the following month in every shop he has been in the past;
2. If this probability is high enough (>50%), predicting the amount of money spent in every specific shop;
3. predicting the number of transactions that he will make;
4. and, finally, forecasting the days in which the customer will make transactions.
Download full document
Our passion towards these topics and the satisfaction in getting such good results led us to write and publish our learning path, our analysis and all the experimentations we have done so far without any exception, hoping that this will help those people who are studying these topics or are interested in such solutions.
To download the full document, simply request it by filling out the form below.