Synthetic data has emerged as one of the most talked-about innovations in modern technology. As artificial intelligence and machine learning (AI/ML) workflows expand across global industries, enterprise technology leaders view synthetic data as a critical capability for maintaining data privacy while accelerating software development. However, a comprehensive 2026 survey of 518 global enterprise technology leaders conducted by Delphix reveals a sharp contradiction between market perception and actual enterprise adoption.
While synthetic data is widely perceived as the ideal approach for protecting data in AI/ML environments, real-world usage trails far behind expectation. This growing disconnect points to a distinct “synthetic data market gap”—where legacy synthetic data tools fail to meet the rigorous quality, realism, and referential integrity demands of enterprise software engineering and agentic AI development.
A Striking Contradiction: Mindshare vs. Adoption
The survey findings highlight a clear paradox in how enterprise leaders view and deploy synthetic data. When evaluating methods to protect sensitive information within AI/ML workflows, 56% of overall enterprise leaders ranked synthetic data as the single most purpose-fit approach—outpacing traditional solutions like static data masking. Among Analytics leaders (data engineering and data science professionals), that preference rises to 59%.
Yet, despite this strong endorsement, practical implementation remains strikingly low. The survey found that 51% of organizations are not using synthetic data in AI/ML workflows today. This includes 23% who have never used it, 15% who experimented but stopped, and 13% who evaluated synthetic solutions but determined they did not meet enterprise requirements. Further, only 28% of organizations report using synthetic data extensively as a core part of their AI/ML operations, and 66% of software engineering and testing leaders report that 66% of their organizations aren’t using synthetic data tools or approaches.
This contrast demonstrates that while enterprise leaders clearly recognize the promise of synthetic data, adoption is hindered by the limitations of existing commercial tools.
The Root Cause: Enterprise Requirements Fall Short
Why is there such a significant gap between perception and practice? The survey data reveals that legacy synthetic data solutions fall short on the two foundational pillars of test data management: data realism and referential integrity.
Enterprise software testing requires data that reflects complex, multi-system business logic and maintains precise relationships across databases and applications. However, according to the report, survey respondents rated legacy synthetic data tools quite poorly across these essential criteria:
- Only 38% of leaders believe available synthetic data solutions provide sufficient data realism.
- Only 34% of leaders state that synthetic data delivers necessary referential integrity.
- Just 12% to 13% of leaders view synthetic data as the most purpose-fit solution for software integration and unit testing.
Legacy synthetic data generators traditionally operate as isolated point solutions or rule-heavy platforms. They require extensive manual configuration—defining schemas field-by-field and mapping complex relationships by hand. As enterprise systems become more interconnected, maintaining these manual configurations becomes unsustainable.
The Agentic Era Demands a Next-Generation Approach
The need to bridge this market gap is becoming increasingly urgent as organizations transition toward AI-assisted and agentic software development. In agentic workflows, autonomous AI agents construct, test, and iterate on software at unprecedented speeds. To function effectively, these AI agents require realistic, scenario-specific data on demand without violating privacy regulations or compromising compliance.
To support this shift, next-generation synthetic data solutions must evolve beyond standalone generators. Modern enterprises require a unified platform that combines AI-powered automated discovery, automated data masking, and governed synthetic data generation capable of operating at agentic speed.
Key Takeaways for Enterprise Leaders
Closing the synthetic data gap requires aligning technology capabilities with actual enterprise priorities, such as requiring data quality to drive decisions. These software engineering and test leaders rank high-quality, realistic test data as both their highest priority and their greatest operational bottleneck. Another requirement, for referential integrity, calls for end-to-end testing across an organizations’s application suite and cross-system relationship preservation. Finally, AI-first discovery and governed data delivery are necessary to keep up with agile development and AI agents.
By addressing these core requirements, enterprise leaders can move beyond legacy limitations and harness high-quality, privacy-safe synthetic data to accelerate software delivery and drive AI innovation.
SD Times Q&A
What percentage of enterprises are not using synthetic data in AI/ML workflows?
According to the 2026 Delphix survey of 518 enterprise technology leaders, 51% of organizations are not using synthetic data in AI/ML workflows today. This breaks down as 23% who have never used it, 15% who experimented but stopped, and 13% who evaluated synthetic solutions but found they did not meet enterprise requirements.
Why do legacy synthetic data tools fail enterprise software testing requirements?
Legacy synthetic data tools struggle with two foundational pillars of test data management: data realism and referential integrity. The Delphix 2026 survey found that only 38% of leaders believe available solutions provide sufficient data realism, and only 34% say synthetic data delivers necessary referential integrity. These tools typically require extensive manual schema configuration and do not scale well across interconnected enterprise systems.
How does synthetic data compare to data masking for protecting sensitive data in AI/ML pipelines?
In the 2026 Delphix survey, 56% of enterprise technology leaders ranked synthetic data as the single most purpose-fit approach for protecting sensitive information in AI/ML workflows, outpacing traditional solutions like static data masking. However, actual adoption remains low, with only 28% of organizations using synthetic data extensively as a core part of AI/ML operations.
What synthetic data capabilities do agentic AI development workflows require?
Agentic AI workflows—where autonomous AI agents build, test, and iterate on software at high speed—require realistic, scenario-specific data on demand without violating privacy regulations. This demands unified platforms combining AI-powered automated discovery, automated data masking, and governed synthetic data generation capable of operating at agentic speed, rather than isolated point solutions.
What is the synthetic data market gap and why does it matter for software engineering teams?
The synthetic data market gap refers to the disconnect between high enterprise mindshare for synthetic data and low actual adoption. Despite 56% of leaders viewing it as the top data-protection approach for AI/ML, only 28% use it extensively. For software engineering and testing teams specifically, just 12–13% view synthetic data as the most purpose-fit solution for integration and unit testing, pointing to a critical tooling maturity gap.
