Synthetic data (SD) has surged in recent months - even data-rich enterprises are testing it. But it isn’t the right fit for every use case.
Why synthetic data is gaining traction
- Data scarcity & privacy: High-quality labeled data is hard to obtain; SD promises scale without PII. Market studies project very fast growth for SD generation tools (30–46% CAGR) as AI adoption widens. Grand View Research+1
- Analyst signals: Prior Gartner research/surveys have argued that synthetic data will increasingly power AI/analytics development this decade (often cited as “majority by 2030”). evidence-hub.aetion.com
- Industry dynamics: With real human-generated data becoming a bottleneck, many teams are exploring SD or hybrid datasets. Business Insider
At the same time, researchers warn about recursive training on synthetic corpora (model-on-model drift, bias amplification) if guardrails are weak. WIRED
Does synthetic data make original KG data obsolete?
I don't think so. Knowledge graphs earn their keep by providing context + causality + provenance—things SD typically cannot supply by construction.
- Provenance & explainability: KGs encode who/what/when and the why via typed, directed edges (e.g.,
TRIGGERS,MOTIVATES,HINDERS). You can trace a claim to sources. SD can be statistically plausible yet source-less, hence hard to audit. - Exclusivity: First-party KG facts are a differentiated asset; many teams generate SD from similar public priors, limiting strategic uniqueness.
- Reliability over time: With KGs you can version, re-score, and retire edges as evidence changes—reducing drift.
- LLM grounding: Graph-aware RAG works best with paths + provenance, not generic synthetic narratives.
Bottom line
Synthetic data accelerates development; knowledge graphs create trust. The biggest wins come when you let SD boost iteration speed while your KG remains the source of truth for explainable, evidence-based decisions.
If you’re curious how this plays out for customer context, here’s my short post on why original qualitative data is non-negotiable: Why original qualitative data is non-negotiable. There’s a lot of buzz around synthetic data as it promises a shortcut to faster insights. But when the stakes are high and you need to make… | Eckhart Boehme