Field Note
Data·20 July 2026·5 min readSynthetic data, real judgement
Where synthetic profiles help, and where they quietly mislead.
Synthetic data has quietly become the default way teams prototype segmentation and personalisation before real data is available. It's genuinely useful and genuinely easy to over-trust. This piece is about knowing which is which.
The executive problem
Synthetic data lets teams move fast: build segments, test personalisation logic, and demonstrate value before a single piece of real customer data is touched. The risk is that it's easy to forget it's synthetic once the outputs start looking plausible.
Why it matters now
Privacy constraints, data-sharing agreements and slow data-platform timelines mean more teams are prototyping against synthetic profiles for longer than they intend to. What starts as a two-week placeholder can end up quietly informing decisions six months later.
Why the conventional response stalls
The usual safeguard is a disclaimer: "this is illustrative, not real." Disclaimers don't survive contact with a stakeholder who likes what they see. Once synthetic output looks convincing, the burden of proof quietly shifts from "prove this is real" to "prove this is wrong" and that's backwards.
The LoweLabs point of view
Synthetic data is a tool for testing structure, not for validating substance. It's excellent for proving a segmentation model or a personalisation pipeline works mechanically. It's a poor substitute for evidence that a specific segment or offer will actually perform, because it can only ever reflect the assumptions it was built on.
“Synthetic data can prove your pipeline works. Only real data can prove your idea does.”
Keeping synthetic data honest:
- Label synthetic outputs clearly and permanently in every downstream artefact, not just the first one.
- Set an explicit expiry, a date by which synthetic data must be replaced with real data before decisions can lean on it.
- Validate structure with synthetic data, but validate substance "will this actually work" with real data or a live test.
- Treat a synthetic result that looks "too clean" as a signal to check assumptions, not a signal of quality.
Founder, LoweLabs
With more than 25 years’ experience across consulting, technology and industry, he has built and scaled multiple Data & AI practices, led complex transformation programs, and worked with some of Australia and Asia’s largest organisations. An INSEAD Global Executive MBA graduate, he combines commercial leadership, deep technical expertise and a practical understanding of how emerging technology can be turned into measurable business value.
What separates a pilot from a product
The gap between a working demo and something people rely on.
Point of ViewEvaluating agents like you evaluate people
Why AgentOps borrows more from performance management than from software testing.
Point of ViewWritten at the Wrong Altitude
Why your AI strategy keeps expiring, and why the next rewrite will not fix it.
Point of ViewThe Answer Is in the Portfolio
Eight figures on AI pilots, nothing on the P&L. Why "double down, redirect, or stop" is the wrong question.
Point of ViewNobody Should Own AI
The turf war between your CDO/CIO, CTO and CFO isn't a governance failure. It's a category error.
Point of ViewDesigning the AI Op Model
Before an organisation can delegate work to machines, it has to understand what it is delegating.
Next steps
Relying on synthetic data longer than you planned to?
We help teams work out where synthetic data is genuinely useful, and where it's time to bring in the real thing. Let's talk.