Synthetic data, created from real datasets using AI, speeds accessibility of production data
Hazy, which specialises in synthetic data, has carried out a proof of concept (PoC) with Vodafone Group R&D and Vodafone UK, trying out synthetic data for training and testing machine learning models to manage customer value.
Machine learning models were trained using large amounts of customers’ data. Vodafone Group R&D team used internal metrics and those built into Hazy’s software to evaluate the synthetic data they created. The team concluded the synthetic dataset preserved the performance of the machine learning model, saving the team time, cost and in some cases “even” improving the model’s performance.
The PoC went live in eight weeks, during which the Hazy team worked alongside Vodafone’s team so it could generate synthetic data itself. The complete training phase of the data generator took one minute: using established methods, it could take as long as several weeks to access production data or create a test dataset manually.
After the PoC
After the PoC Vodafone Group is now considering the best ways to use synthetic data capabilities, identifying and and prioritising future use cases for various types of synthetic data.
Luke Ibbetson, Head of R&D at Vodafonesaid, “We are really excited about the potential of Hazy Synthetic Data. Our team was impressed with the quality, utility and usability of the generated datasets, as well as the software’s ease of deployment.
“We are keen to explore the power of a synthetic data platform for Vodafone teams to access useful synthetic data. This software could be a powerful, cross-function tool, saving time, increasing productivity of data science teams, accelerating project deliveries, and enabling us to operate more efficiently.”
Harry Keen, CEO at Hazy,said “Thanks to the close working relationship developed with the Vodafone team and a great choice of data set for customer value management, we were able to rapidly develop a synthetic version that met the success criteria for privacy, speed and fidelity.
“We are looking forward to building on this success to operationalise the value of synthetic data across the group.”