ING Belgium, part of the Netherlands-based bank ING Group, claims to have spent the past four years building one of Europe’s most advanced synthetic test-data capabilities, driven by AI, trained on production-like datasets, and designed to solve what many banks still call their “number one” testing bottleneck: reliable, timely and privacy-safe test data.
Speaking to the challenges shared across most QA organisations, Wim Blommaert, Head of Test Data Management at ING Belgium, put it plainly: “Testing is only as good as the available test data.”
He added that “the struggle, mainly due to ever-increasing privacy regulation, to get test data, including the special cases, when we need it, where we need it, we have all been there.”
Blommaert recalled demonstrating two payment-screen samples at a recent conference, one supposedly real, one supposedly artificial. “The vote was split 50/50. However, both screens are synthetically generated by AI.”
He disclosed the realism even caused friction during early deployment at ING: “We got some pushback from the testers as they could not easily make the distinction with the ‘real’ payments they had created.”
ING ultimately added a “TEST” stamp just to avoid confusion, Blommaert wrote in a recent ING blog.
The maturity of the models pushed ING to scale the approach. “Today we have +20 applications for which we have generated synthetic test data. We are building a platform.”
Blommaert outlined four core challenges with traditional approaches: cost, slowness, privacy constraints and incompleteness.
He noted that copying production data is expensive: “Copying data from production comes at a cost.” Teams also struggle with access: “On average people had to wait three weeks between the time they requested the data until it was loaded to the test environment.”
On privacy, the barriers were even more pronounced. “Sometimes data cannot be copied from production, even if masking is applied.” Masking itself often introduced new risks or inconsistencies: “Masking can also turn out to be more complicated than anticipated.”
And production data is rarely sufficient for edge-case testing. “Special or new cases might not exist. Synthetic test data can help to solve these problems”
For Blommaert, AI-driven synthetic data was a natural answer: “Basically, it is automation, so once set up, it can be very fast and save you time and effort.”
And crucially for regulated financial institutions: “Synthetic data is very privacy-safe and under GDPR it is not considered personal data.”
“You can force the AI to generate data specifically by following your instructions to test special cases.”
– Wim Blommaert
Blommaert emphasised the enterprise-wide impact of solving this problem: “Thousands of engineers at ING can benefit from this.”
Describing how ING applies the method, Blommaert explained: “What generative AI can do for images, it can do so for tabular data as well.”
In the bank’s regulatory reporting project, ING validated synthetic tables by comparing distribution patterns: “Frequencies and distributions of columns should match.”
Even quality issues can be reproduced where useful: “It could be… that this data quality issue also existed in the real data, and it got learned by the AI. This can be useful for testing negative test cases.”
If errors are not desired, ING simply instructs the model: “If we don’t want these ‘errors’… we can ask the AI to always respect the date sequence.”
Data vault
ING’s implementation is built on MIT’s Synthetic Data Vault (SDV), as Blommaert explained: “At ING we are using Synthetic Data Vault (SDV), a software originally developed and open sourced at MIT.”
Model options vary, but ING largely prefers statistical approaches: “As the GAN approach does not scale for bigger models and is very data-hungry, we focus mainly on the copula.”
Training sets can be surprisingly small: “Training sets can be small, having 20K rows is typical, but can then be used to generate millions of rows of synthetic data.”s
One of the first major use cases landed in ING’s SEPA payments testing. Blommaert said: “We took a real set of 5,000 payments from the production queue, trained an AI model and generated 10,000 synthetic payments in less than 2 minutes.”
The team then injected these into the Acceptance environment, but hit a validation issue: “The back-end did not recognize the synthetic account numbers.” ING solved this by combining techniques: “We chose the latter and replaced the synthetic IBAN with the IBAN from the Acceptance environment.”
More recently, ING tightened its test coverage by moving from large random sets to targeted precision: “Instead of generating 10,000 random payments, we generated only 20 payments based on the actual test cases.”
The model can even invent unseen edge cases: “We can even ask the model to generate values that it has not seen in real data… but that might still be interesting edge cases.”
Automating mock API mappings
Another breakthrough came when ING applied synthetic data to API mocking.
Normally, mocks require manually writing 100 or more request–response pairs. “Filling out 100 of them will take a lot of time and lead to errors.”
By mining existing API logs, ING changed the game. “We used 2,000 of these example pairs to train the model. From this model we can then generate the pairs we want to feed to the Mock.”
Blommaert called it “a nice example of the combination of two technologies (Mock and Synthetic data) to help engineers.”
Blommaert concluded ING’s four-year journey with a simple reflection: “We have shown that access to good test data is often a problem and that synthetic data generated by AI can be a solution.”
The bank is continuing to expand its enterprise platform and encourages teams across the industry to explore the MIT-backed toolkit.
As he put it: “If you want to continue this journey with us, you can have a more technical look by trying out the open-source version: SDV.”
Why not become a QA Financial subscriber?
It’s entirely FREE
* Receive our weekly newsletter every Wednesday * Get priority invitations to our Forum events *
REGULATION & COMPLIANCE
Looking for more news on regulations and compliance requirements driving developments in software quality engineering at financial firms? Visit our dedicated Regulation & Compliance page here.
READ MORE
- Leapwork engineering head: Why test automation so often fails to deliver
- World Economic Forum warns financial sector must strengthen AI risk controls
- How Banca Progetto is hard-wiring quality into Italy’s digital banking space
- How Wealthsimple builds quality into the product, not around it
- Digital revamp puts spotlight on internal controls at China Construction Bank
WATCH NOW
QA FINANCIAL PODCASTS
Listen to Sudeepta Guchhait on Nasdaq’s new Mimic AI testing platform
QA Financial sits down with Sudeepta Guchhait, Senior Director of Product Framework & Quality Engineering at Nasdaq
——–
Listen to Wesley Scheffel and Robin Rain on Schroders’ DevOps strategy
We catch up with Wesley Scheffel, Head of Cloud Platform and Product Engineering at Schroders, and Robin Rain, Head of Cloud Platform Architecture
——–
Listen to Citi’s Jason Morris on Lightspeed and the future of continuous delivery
Jason Morris, Head of Developer Pipelines for Securities Markets and Banking at Citi, talks about Lightspeed
