Only 34% trust synthetic data’s referential integrity — and it’s holding adoption back
The New Stack Mayank Ahluwalia
Only 34% of leaders trust synthetic data to keep related records linked. That trust gap is why adoption is still stuck, even as AI raises the pressure.
Based on reporting by The New Stack, Mayank Ahluwalia — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Enterprise leaders clearly like the idea of synthetic data. The problem is that liking the idea isn’t the same as trusting the output. Perforce Delphix research says the biggest blocker to adoption isn’t awareness, compliance, or governance. It’s data quality, and the numbers are blunt: only 38% of leaders say synthetic data is realistic enough, and just 34% believe it preserves referential integrity.
That matters because developers don’t care much about the theory. They care whether the data helps them test real systems without exposing sensitive information. Yet 66% of software leaders say they aren’t using synthetic data in development and testing. If the generated records don’t keep the right relationships intact, teams end up testing fragments instead of real business processes.
The source makes a simple point: synthetic data can’t just look plausible in one table. Enterprise apps spread information across databases and systems, so customer records, orders, and payments have to stay linked. If those relationships break, testing can produce false confidence, defects, failed releases, and bogus results. Generating a random number that fits a field is not the same thing as preserving the logic of the system.
There is some adoption already. More than half of enterprises, 51%, use synthetic data to protect sensitive information, often alongside masking. But the article argues that broad use will only follow if tools can consistently preserve realism, generate data in dependency order, and keep up as schemas change. AI makes the issue sharper, not softer: code generation is speeding up, but data access is not. If software delivery is moving faster, the data has to keep up too.
The larger message is pretty plain. Synthetic data is not being held back by hype fatigue. It is being held back by trust, and trust here means whether the data behaves like the real thing.
My take — AI-written commentary, not fact-checked reporting
This is the right fight to pick. Synthetic data without referential integrity is just expensive cosplay for databases, and engineers can smell that from across the room. The industry keeps selling speed, but the real bottleneck is still whether the data behaves like reality instead of a cheerful lie.
Read more about this at: The New Stack