Introducing MOSTLY AI 2.2 - now smarter, faster and more efficient
Generate now!

Synthetic test data for digital banking

Download the case study


Smart, data-driven
product features,
such as account
balance prediction
and responsive UX
development sprints
by several days as a
result of near-instant
availability of
synthetic customers
transaction data.
Demand a highly
realistic product to
internal stakeholders,
earning valuable
support throughout
the organization.
Shared granular, yet fully
anonymous synthetic
data with several external
partners to develop
further products
and services, such as a
bill splitting app, born out
of identifying genuine
customer needs.


One of the largest retail banks in Europe developed a mobile banking app that aims to be a true alternative to in-person banking. To provide a high-quality user experience, extensive testing of the app with clients’ transaction data was crucial. However, the bank’s IT department had to heavily mask transaction data due to privacy policies, and the resulting test data failed to provide realism in transaction amounts, dates, and so on. Dummy data could never match the smart synthetic dataset’s granularity and realism, failing to provide the complexity necessary for testing a product so important to work flawlessly. Imagine you could create realistic customers at the push of a button!


MOSTLY AI's synthetic data platform, our synthetic data generator was delivered to the bank through a REST API. The algorithm - fed with raw client data - learned its patterns and properties. Once the algorithm was trained, any number of new, realistic synthetic users could be generated. Using a smart test data dashboard, the product development and testing team could generate synthetic customers based on pre-defined parameters, such as the number of accounts, income range, urban or rural address, and others. New, predefined synthetic customers could be created using behavioral data generation to test edge cases.


How can you tell that your test data is exceptionally high quality? We decided to test the testers and presented a mix of synthetic and real customers, using the MOSTLY team’s real banking data. Product developers couldn’t spot synthetic customers in the batch - that is how realistic and high-dimensional the smart test data was. Generating smart test data took significantly less time than anonymizing the initially used and discarded dummy data.